Wireless Microphone Troubleshooting

Took me a moment (but I just woke up, so that's my excuse).

The Source pin should be connected to the "ground" terminal of the JLI2555. Electrically your current setup looks like this:

When you connect the Source/Ground like this:

You get a complete circuit. With the ground pin "floating" there's no circuit from Source to Gate, so charge won't drive the JFET's gate - current can't flow through fresh air unless you use a Tesla coil*! 🙂

Give that a try and drop us a note and we'll take it from there.

* I'm reminded of a talk by the great Rick Hartley here. In that he says that energy is in the fields (electric and magnetic) not in the copper. It's a strange concept to grasp, largely because it goes against how we think of electricity generally. But everything from the mic and television all forms of light are wireless - energy sources. We use copper and other conductors to steer the energy which we see as current and voltage.

Fresh air is a very good insulator and requires thousands of volts of potential difference to cause the gases in air to break apart (ionise) and allow current to flow. That's what causes those huge sparks we associate with Telsa coils.  

@marcdraco 

Thank you for your insight! I’ve tried this kind of setup when I stumbled upon the Furball Mic project by DJJules (the creator of OPA Alice), where he connected the source and ground pads. When I tried it myself, there was still no sound captured, just noise, and I was on the brink of giving up.

Then I decided to troubleshoot it by asking ChatGPT (not always reliable, but better than nothing haha). It told me that if I want to modify the JLI2555 to behave like a universal electret mic (like the JLI2590A), I should place a high-value resistor between the gate and ground of the capsule, around 2.2MΩ to 4.7MΩ.

I only had 1MΩ resistors, so I soldered three in series and hooked them up to my Blue Yeti microphone. (Side note: I accidentally soldered the lavalier’s battery in reverse earlier and broke it. Rookie mistake haha.) Now I’m finally able to capture my voice, but there’s some buzz or hum in the background, probably due to how I arranged the components at the backside of the capsule.

Maybe next time I’ll try to make a PCB for it, but I’m still learning how to use EasyEDA.

(I can't attach any media as there is a warning below that I can't have any attachments for today T^T)

The resistor shouldn't really be necessary. If you are using one it needs to be in the 1 Gigaohm+ range or the extra load created by the resistor will take most of your signal.

You can make a very high value resistor with on some spare board by drawing a line in graphite - pencil "lead" and that will be more than sufficient.

The weird thing is that electrets should work without the load resistor, it's only 100% necessary for an unbiased capsule.

Come to think of it there is a way you can use a "small" high value resistor but that requires feedback from a voltage divider at the source terminal so that's a bit of a faff on. When I tried this technique myself I was always annoyed by the amount of background hiss. 

The idea is to make the voltage at the the source equal to the voltage at the gate, something like this:

The idea (called bootstrapping) works like this.

The basic circuit is a source follower so the voltage at the source "follows" the voltage at the gate exactly - a small drop due to the the FET's transconductance. A high gm (high gain) JFET is essential here because the voltage dropped by the Gate-Source junction determines how large the resistor will appear.

The voltage at the source terminal is raised to a voltage sufficient to bring the gate slightly negative with respect to the gate - this happens entirely due to the JFET operation. But the clever bit is in the feedback. With this arrangement the voltage at the bottom of the 1Meg resistor is very close to the voltage at the top, as seen here in the trace from LTSpice. The green trace is the input voltage at the Gate and the purple one is the Source. 

Since resistance = Voltage/Current, very little current flows from the Gate via the 1Meg and 1k resistor making it appear much larger. The capacitor (typically 10 - 100 n), feeds AC at signal frequencies back to the bottom of the 1 Meg resistor. The capacitor is necessary to block the bias voltage at the source from reaching the gate.

In simulation this gives (around) 600 mV drop across the 1 Meg resistor with just 2 pico-amps (pA) of current flowing so from Ohm's law that makes the 1M resistor look like a 1G load.

@marcdraco

I forgot to mention that when I connected the JLI capsule with a 3MΩ resistor and a 2SK30A to the Blue Yeti, I didn’t use the LM386 preamp board. I totally forgot, haha. Should I try connecting the capsule circuit to the preamp board first, then to the Blue Yeti?

@renzevan How are you connecting to the Blue Yeti?

I don't know anything of the internals of that mic. Don't suppose you could drop a pic so I can take a look?

I've some experience with other chips similar to the on used in the Blue Yeti and they have their own mic pre-amp so you shouldn't need an LM386 to amplify the signal from the capsule.

@marcdraco

I'll drop the pictures in PM, as I can't upload any media as of the moment here :((

It's an anti-spam measure I expect and this is still quite a modest site in terms of actual scale as most of Matt's stuff is handled by YouTube.

@marcdraco 

I’m now able to upload media attachments, so I’ll start by showing what I’ve done with the capsule and how I arranged it.

This is how I soldered all the components at the back of my capsule. It’s a real mess, but that’s how I do it when I’m just testing things hahaha.

Here’s the part where I connect the source and drain of the capsule. Since the Blue Yeti has three capsules (blue and red for the left and right drains, and both black for the source), I didn’t use the rear socket. I just combined the drain wires, did the same with the source wires, and soldered them to the pins of the JFET.

This is the sound of the capsule connected to the Blue Yeti after recording. It has a lot of noise, and note that I forgot to use the LM386 preamp, so the capsule is directly connected to the Blue Yeti.

It's the lack of a screen around your twisted pairs. That's 100 Hz mains. Screen the capsule (do that first - it's the worst offender) and most of that will go away. A little will come in via the wires too. 

As a matter of some weirdness - the JLI/TSI capsule is very poor in this regard - it's designed to sit inside a Farday cage. But the clones are more sensitive to sound pressure and don't suffer with the hum as badly. Probably due to the construction but don't let that persuade you, the JLI capsule is WAY better.

@marcdraco 

Regarding placing the capsule in a Faraday cage, do I need to strictly follow Matt’s design where he used solder wick as shielding for the wires? I’m concerned that using the shield and soldering it to the ground tab might cause issues, since the capsule’s ground is also connected to the source pin of the JFET. Also, where should I connect the other end of the solder wick on the Blue Yeti’s PCB? Will any ground pad on the PCB work?

Faraday cages - strictly - are not connected to ground but in microphones it's usual (due to the cost of cable) to use "screened, stereo cable". Matt's version was simply a rather clever way to maintain that beautiful retro aesthetic which I fell in love with!

One way to maintain a true Faraday cage is to enclose the capsule in a brass shield (steel works too) but don't allow the shield to actually connect to the capsule. This shield would connect to the braid but you put a third wire from the circuit GND reference to the capsule ground tabs.

All three inner cables can be gently twisted together or you can twist just the signal pair and lie the ground alongside it. Another option which would be fiddly would be to get a fine screened stereo cable (seven threads per insulated core) and pass that through some large solder wick. The idea of both systems is to allow the interference to get trapped by the shield will keeping the interference away from the ground wire.

It's not really necessary to go to all this trouble (the cage works fine connected to ground) but that's not required for it to work. We just use it to double up as a ground reference.

I expect the Yeti (which has three low cost capsules) is running them in "Common Source" configuration so the designer will expect the Source pin to be internally connected to the FET's case. There's not a huge advantage to go to the extra trouble of a differential input if the entire thing is screened as it is in the Yeti. Differential does give you (roughly) a 3dB boost in sensitivity but that's more conveniently obtained with a little extra gain at the amp.

I'm curious to know what the IC is on that input board. It looks like the digitiser is mounted on a separate board underneath there. I would assume that visible IC is a dual op amp but without knowing the part number it's impractical to guess. The Yet is designed down to a price, despite not being that "cheap" in real terms. They've done this to tempt people in with the multiple patterns at the cost of quality. 
  

@marcdraco 

MAJOR SOUND UPDATE!!

(Not a perfect output yet, but there’s definitely an improvement haha.)

I’ll talk about the recorded sound later. Just dropping some updates first.

I was only able to capture the op-amp (IC) used in the Yeti’s preamp board since I already reassembled it. Here’s what it looks like.

Also, just an extra detail: this is the capsule that the Blue Yeti uses.

As for your thoughts on the Blue Yeti’s sound quality, you’re absolutely right. They sound terrible for their retail price. They pick up a lot of background noise, the pickup patterns works, but in my opinion, are just gimmicky, and the audio quality is poor. The mic isn’t even balanced well. If you loosen the side screws a bit, the main mic swings and hits the metal stand, which could dent the capsule mesh. I only have this mic because someone sold it to me for around 15 bucks, which is fair for the performance it gives haha.

Now, for the tests and mods I’ve done:

• I removed the 3MΩ resistor, like you suggested, since based on what you discussed, wasn’t really necessary.

• I tried connecting the capsule and JFET setup to the Blue Yeti using the LM386 preamp, but it didn’t pick up any sound, just a loud hum noise.

• Since the noise was more of a hum than a static hiss, I assumed it was interference due to lack of shielding and the use of long wires between the capsule, preamp, 5V line, and Yeti board.

To alter the old setup, I soldered the JFET pins directly to the LM386 board without using wires and placed the whole capsule and preamp inside the Yeti’s mesh grille. I still powered the preamp with USB since I couldn’t tap voltage from the Yeti board (scared to ruin a classic mic haha).

I didn’t use solder wick shielding for the mic wires yet because I’m still unsure where to ground it. Maybe I’ll solder one end to the ground tab of the capsule and the other end to the Mic- pad on the Yeti, since the capsule’s ground and source are already connected.

Then I tested the setup by recording. I kept my expectations low to avoid disappointment, but to my surprise, it worked. My voice was picked up without clipping. There’s still some interference, but it’s a big improvement. My first recording (which I attached before) had terrible clipping every time I spoke, but now my voice is clear even though hum noise is still present.

Yeah, the JLI capsule is superior. Those capsules (not sure the manufacturer) are pressure cardioids but presumably low quality. 

Logitech could (and should) have made a better job of that layout. I'm a little shocked to be honest.

At the price you paid, it's not a bad job although I've heard a lot of people reporting the mics blowing up. It's not clear what is going south so I'll try to find a busted on eBay and see if the V2 can be used as a potential upgrade. I'm in the process of splitting it into two boards, not out of necessity but as a way to ensure if one section (which is quite mature) works and the other (experimental, due to the previous IC going out of production).

The only way to isolate hum (which is inductive interference from the magnetic field) is to shield all the high-impedance paths with a Faraday cage (which you can connect to any ground. To avoid ground "loops" you should take the ground as close as you can to the USB receptacle. 

This is called a ground Mecca and while it's only really a bit of a crutch, it's often the only way to get a quiet ground.

@marcdraco 

When it comes to shielding all the high-impedance paths, I’m thinking about how Matt used solder wick to shield the wires. Am I correct in assuming that he did this because his mic setup was open? If that’s the case, can I just replicate his shielding by placing all the components inside an old microphone body that has a good connection between the metal mesh around the capsule and the main metal housing?

Yup.

The thing that's hardest to understand with Faraday cages is that they have to be continuous wrap around the object we're trying to screen - hence the term "cage". They don't have to be connected to a reference point. The wonderful Rick Hartley points this out from the time he was designing boards for TCAS (tactical collision avoidance systems) for commercial aircraft.

The skin of a modern airline is a Faraday cage and that's no accident and it's why it's possible to get hit by millions of volts of lightning and for the passengers not even notice so much as a tingle. The current runs around the outer skin of the aircraft and just carries on to Earth.

Any metal bodied vehicle does the same on the ground so if you are caught in a particularly bad electric storm, you're far safer staying in your car than running for (for example) some trees or other shelter.

To explain fully it's necessary to understand electric and magnetic field flow but without diving into that, all you need to know is the the energy follows the path of lowest impedance to ground - quite literally mother Earth in this example. At the effective frequency of a lightning burst which is difficult to quantify, but think in terms of MHz (it's the time it takes to travel from the cloud base to the Earth) which means that the current flowing in the metal flows harmlessly around the outside of the metal to a depth of a fraction of a millimetre.   

All that matters is that the cage is continuous (so far as the fields are concerned) which is why the cage can have more holes than a poorly written movie. The actual size of the grid can be taken from the 1/2 the wavelength of the frequency of the radiation, specifically the magnetic field we're trying to keep out.

When the cage is put into a magnetic field, it induces a current in the metal which is lost to the air as heat. We don't feel it of course because the amount of energy is tiny, but that's enough to harmlessly dissipate the interfering signal. 

Mics like ours need power so we can make the cage double up as a ground reference which is fine on a microphone, but on the aircraft, a grounded shield could easily conduct energy from a strike into the electronics and blow things! (In days gone by, televisions and radios would occasionally be destroyed by lightning hitting the outdoor aerials and doing just that.) Very tall structures always have lightning rods for the same reason.

So in a nutshell, placing the mics inside the body of the Yeti should do the trick!

@marcdraco 

Thank you for that confirmation! Now I do feel a little bit smarter about this topic than I was before, HAHAHA. Now I’m just waiting for the rear capsule PCB I custom-ordered from JLCPCB, so I can make the components behind the capsule look neater. I’m just hoping my trusty metal ruler measured the dimensions accurately, since I’m really new to PCB making, yet I already went ahead and custom-ordered one right after learning and making my first PCB (a "great decision on my part" HAHAHA). I’ll also try to upload my PCB here once I confirm that it fits inside the 2555 capsule.

Brilliant stuff - but as the old saying goes, "a journey of a thousand miles begins with one step (onto the plane)".

I added that last bit myself because no one really wants to walk 1000 miles. Even the Proclaimers.

Posted by: @renzevan

↑

@marcdraco 

 Now I’m just waiting for the rear capsule PCB I custom-ordered from JLCPCB, so I can make the components behind the capsule look neater. I’m just hoping my trusty metal ruler measured the dimensions accurately, since I’m really new to PCB making, yet I already went ahead and custom-ordered one right after learning and making my first PCB (a "great decision on my part" HAHAHA). I’ll also try to upload my PCB here once I confirm that it fits inside the 2555 capsule.

Yeah, I know those feels. First couple I did I didn't get the slots quite right. If yours doesn't fit, please do feel free to take my tested ones from GitHub and alter the components as you need. At least you can be 100% sure that everything is measured correctly (and that there is sufficient space for the back of the capsule to breath or you'll produce an Omnidirectional mic rather than a cardioid one. Matt ran into this head first but his solution was rather nicer looking than a simple board and rather adds to the aesthetic.

@marcdraco

Actually, I slightly copied the design of Minima, specifically the dimensions. I just eyeballed the distances between the holes because I don’t know how to import your design into EasyEDA. The pin configuration of my JFET (2SK30A-GR) is also different from the 2N4416 metal can type, so I used a different component footprint. I also connected the source on the PCB to the ground tab. Overall, it’s still based on the Minima, which follows what Matt told us that we shouldn’t block the rear side of the capsule; otherwise, it would affect the sound and make it perform differently from how it’s intended.

I also designed it so that I have the option to use a resistor or not (though I won’t be using one anymore, as I’ve decided to follow your suggestion).

(I just named it after my girlfriend so she won’t ask how much I spent on this project haha.)

Brilliant, the whole point of Open Source hardware is you can build it, make it, improve it... all without cost. So that makes me very happy.

And naming it for the girlfriend? A master stroke, sir, I salute you! I'm way over budget but I'm having so much fun, I'm like a kid in a candy store with their parent's American Express card. 🙂 

The tabs are exactly 20 mm apart (10 mm centres from the centre pin). I know this because I got it wrong once and that was painful. I got the measurement from the original CAD drawings but in my haste I'd nudged them about 0.5 mm closer than they should be and boy, such a small error made it very tight indeed.

I really must try to learn EasyEDA, I can't access those lovely coloured Silkscreens with KiCAD. 

As for that gate resistor, it's a case of having (the space) and not needing it OR needing the part and not having anywhere to put it ... no option really. We can't skip it on a 48V unbiased capsule as that has a DC blocking capacitor in place at the gate and a biased capacitor (the capsule) behind that.

I expect it just works because the gate leaks a little bit of charge into the electret material but one of these fine days I'm going to build something to find out for sure. 🙂

Assuming all goes well, that will be an excellent upgrade path for the otherwise "so-so" Blue Yeti. I tried to find a busted one to see what's going on inside that makes them so unreliable but people are asking silly amounts for them - especially in non-working condition.

In fact, I dare say, we could fit a V2 UUM in the Yeti, although I'll leave that to someone else. My input stops at a working PCB, I'll let others perfect the casing, that's the fun of DIY after all!

@marcdraco 

My PCBs came today!

Luckily, all my measurements were correct! Although the fit is tight, it still fits HAHA. I just soldered it without pushing the PCB too close to the back of the capsule to avoid blocking any holes on the rear part of the capsule.

Now I'm just waiting for my new wireless lavalier to arrive before I set everything up inside my old BM800 microphone. I do have to sand the inside of the mic, though, since it has thick layers of shiny paint that are preventing good electrical contact between the parts of the microphone frame.

I also recreated and printed a THT PCB for the OPA Alice, since I can't afford the ready-made SMD version from JLI Electronics (price + shipping).

But I guess I was a bit too confident in the PCB I designed, because I'm pretty sure there's something wrong with my schematic.

This is the schematic from DJJules,

and this is the one I made.

The part that made me doubt my work was the green arrows in DJJules’ schematic. Since I wasn't very familiar with EasyEDA at the time, I used the arrow that has the ground symbol. Now that I’ve checked my PCB, I’ve ended up connecting all the components with that arrow directly to the ground parts of the PCB, which I think is very wrong.

The reason I decided to make the OPA Alice is because I already have a board based on the one DIY Perks made, but I’m pretty sure the THAT1512 chip on mine is a fake. On the other hand, the components used in the OPA Alice, like the OPA1642, are much easier to source here in our country.

I also want to ask, how can I possibly create a 1GΩ resistor? I’m planning to use a scrap PCB that has two pads spaced far enough to draw a graphite line between them using a Staedtler HB pencil. The problem is, I don’t have a multimeter that can measure up to 1GΩ.

The Alice is very well established and the OPA1642 is an excellent Op Amp. There's precious little we can do at at 48V with only a couple of mA supply current, so the "usual low noise suspects like the NE5532 are non-starters. Which is a shame because they are very low cost for assembly. It''s an op amped version of Jorgen Wutke's famous design for Shoepes from the 1960s with lower noise and much lower distortion through the effects of 100% negative feedback.

Those arrows are ground references - so you have that correct - up to a point (and this is where I think you've stepped on a rake) but it's largely due to a "mistake" in the schematic which should have been corrected years ago. Experts will spot this right away but if you're still in the early stages, it's just waiting to rise up and smack you right on the nose. 

You don't need to measure the resistance of a graphite resistor and it's probably not necessary since the input impedance of a JFET is already staggeringly high (and being an op-amp, it it's got an internally biased differential pair internally anyway). The 1G resistor allows a tiny amount current (pico amps) to bias the gate-channel correctly, but it's really only 100% necessary if you have a 48V bias on the capsule isolated by a polymer capacitor of around 1 nF (the 2555 is an electret so that's not required in most cases).

However, In this case, the designer has applied a bias voltage (see later) via that 1G resistor so it has to be present for anything to work. I'm leery of applying any voltage to the capsule but the fact there are thousands of these floating around and working is testament to the fact it's likely harmless. I suspect it might tension the capsule slightly as it will affect the amount of charge however and that could make it a little "peaky".  I haven't tried this though and such a low voltage is unlikely to have much in the way of an audible difference.

There is an error on your layout, you've swapped out the Zener (12 v) for a signal diode so that won't regulate the power. If you put a P48 phantom in there it will destroy the Op Amp. Also, C1 is a bit too far away from the IC. It's there as a power decoupling and it needs to be as close as physically possible to the chip. You don't need anything fancy, a little ceramic disc is fine. Might work though, I've done it on A/F boards when I was a kid and they generally worked OK.

Which raises the question, why did you think you've done it wrong? Alice is designed for 48V phantom power and it won't operate with anything less than around 20 V DC on both inputs. You'll see a voltage drop across R3 and R4 caused by the load from the OPA1642 and the zenner which may lower the voltage below the 12V. Based on the IQ for two devices (3.6 mA) and the 2 x 2K2 in parallel that runs around 4V drop - which is below the avalance for a 12V zener - so the regulation won't work correctly. The board will still get power but any ripple will be passed directly to the JFET (if you've used on on your version of Minima. In fact, that's not actually required as the OPA1642 has its own.) It all depends on what you're powering the Alice board with, so I'm guessing at this stage. The OPA1642s have reasonably PSRR but you can bet the effect is primarily for low-frequency noise like mains hum, any RF is just going to walk through the front door and pollute the output, in worst case saturating it. (That's another little gotcha the IC manufacturers don't mention...)

The OPA1642 can run happily down to 5V total supply so a 6V zenner will do the job here if you're not able to give it 48V phantom.

The subject of grounding has driven me so far up the walls over the last few years that I could see touch the cloud base on a clear day...

=== this bit gets a bit necessarily technical so HMU if it's not clear ===

Grounding seems (on the face of it) to be so simple... what could go wrong, but in fact it causes more issues with electrical noise and even pickup that it gives many engineers, even ones with decades of experience, serious anxiety.

The problem is the difference between the schematic (where we often tie grounds for convenience per C2, C7 and C8 on yours) and the board where we have to imagine where the field lines (electric and magnetic) are going to end up. Magnetic radiation can bleed several line widths of distance putting noise where it has no place being and electrical fields create little capacitors that are also rakes for the unprepared. Dr. Eric Bogatin (one of the masters) said in a talk that no one becomes an expert without making a lot of mistakes, so you can take some succour in that. I've made plenty myself - some in this thread while I was still getting back up to speed with modern best practise. 

I'm going to moan now... then "open" ground symbol shown at the junction of R8 and R8 is NOT a ground. This is very bad form and although I understand what the original designer was going for this isn't just a rake, it's a whole field of long grass full of them.

That point is correctly called "virtual" ground, it's not a ground at all but a reference point to centre the non-inverting input (pin 3) of the OPA1642 to 0.5 Vcc. So 6V in the Alice (or 2.5V if you working on a lower voltage phantom).

"'ere be rakes"

If you're wondering why it's because you can't drive the input of an op amp below it's negative rail (many are destroyed or have their properties affected permanently, but this is mostly bipolar inputs since the base-collector junction can be burned out. JFETs are a little tougher and with such a tiny amount of current, it won't destroy the device). At signals of a few 100mV max, it's the JFET will probably work without issue - it's something I need to breadboard up just to try.

A single JFET's gate-channel connection is a reverse biased diode so although we tend to think of JFETs as depletion mode (the current in the channel is maxed out when the gate is at 0V) they will continue to operate until the gate potential hits around roughly 600mV (forward diode drop) at which point it will clip as the channel reaches the point at which it's reached maximum impedance. Not something one should rely on though... Internally the JFETs of your op amp are connected as a differential pair so that doesn't apply. 

Since we don't have a negative rail here, we have to bias the input to some value - ideally around 1/2 supply to get the maximum swing at the output. Given there's no overall gain in this circuit (it's a follower which drives the "hot" pin followed by a unit gain inverter driving the cold (out of phase) pin any voltage above a few hundred millivolts will be fine.

The actual ground reference on pin 5 is harmless since Pin 6 is pulled up by the output of the first amp. Any remaining DC is removed by the DC blocking C4 and C5.

Op amps are DC-coupled amplifiers so the DC bias voltage appears at the output (Pin 1) and with a wiggle caused by the capsule and that biases the second amplifier. Note that pin 5 is actual ground - the same potential as the rest of the board. Biasing this to 1/2 Vcc will remove the "common mode" DC bias (+6 V + -6 V = 0 V plus any A/C signal) but that's not really required because the output has capacitors to block any DC.

"'And here's where it all goes wrong..."

If you've tied all the grounds together in one lump, you've effectively shorted out R8 and the bias voltage won't appear - and the device will clip every lower half cycle as the signal goes slightly negative with respect to board ground.

I wish designers (as we should all know better) would stop using ground symbols for bias voltages because it's a certainty someone will stand on the rake and be left scratching their heads wondering why their carefully designed PCB doesn't work.

It's impossible to cram all the usual grounding mistakes into a single post - there are books written about this, although most of the issues really only crop up at very high frequencies of digital ICs (logic, ADCs, DACs, MCUs and so on).

Current at low frequencies - DC to a few KHz - moves through the path of lowest resistance but as the frequency goes up, more and more of the current flows through the path of lowest impedance. Which seems nonsensical until you think it terms of fields and - this is relevant here - the electric field tends to dominate. A PCB trace on layer 1 (top) couples capacitively to the nearest piece of copper and that's almost always the ground reference plane. Note I say PLANE here, don't route your grounds separately. That's a recipe for disaster. 

In fact (and the way I was taught and I've had to unlearn) is we should connect all grounds to a single point on the board. This ensures that all return currents have no choice except to flow through those tracks. Turns out "ground mecca" is a historical thing that, like the human appendix, serves no real function on a modern board and can often make things worse.

Since the fields "live" in the FR4 plastic the sandwich formed between a trace and the plane is a small capacitor so AC current flows "through" the capacitor to the ground plane and follows your trace everywhere it goes. (A/C Current doesn't flow through capacitors, it's just convenient to think of them that way, the effect is caused by the electric field moving free ions around and holding them in place on each plate.) If you have the money to burn on a spare PCB although you can do this with a a thin piece of paper and some aluminium foil at a pinch, you can do this:

You'll ideally have an AC ammeter (or a small resistor of a few 100 ohms and a A/C voltmeter) you can do this experiment yourself. At first glance it would appear that current will flow entirely through the meter on the right and nothing will pass through the one on the left. At DC and very low frequencies (the actual frequency where you can see the effect depends on the thickness of the insulator) the nothing reaches the meter on the left - as it takes the path of lowest RESISTANCE which is the short through the first meter. Ammeters are better since they have a very small resistance at the insertion point).

However, and this fries your noodle when you see it, as the frequency gets higher, the signal starts to move the needle on the left and eventually nothing flows through the meter on the right. At high frequencies (MHz and above)) the electric fields - and this is the key  - couple to the solid plane underneath the track and runs around the long route! While this is generally something that's taught at degree level and only then in high-speed engineering courses, you will find videos showing this is real. Even if you don't follow the logic (and it's anything but intuitive) it's worth knowing because it's crucial if you ever want to work with today's logic chips which have rising and falling edges measured in sub-nano second terms so even if the oscillator only switches once a day, that change in potential causes a very fast edge to blast through the board creating harmonics up the yazoo!

We were also taught back in the 1980s (and this is still common knowledge) that ground pour doesn't hurt. I've stepped on that rake too many times and now I'm working without any pour at all. It's nerve wracking to apply modern best practice with something that I've had drummed into me with a large hammer.

The theory is that a trace on the bottom layer (of a two layer board) will couple to the copper pour on the top but in practise, the pour also picks up induced current (magnetic field) from the tracks that run through the pour and you get even more noise than you otherwise have!

Worse still, since the value of a capacitor is determined by the distance between the two conductors (less distance = greater capacitance) - the trace and the nearest ground plane - the rather large spacing (typically 1.6mm) between the two conductors makes the parasitic capacitor much smaller and increases the frequency of the very signals it's supposed to be trapping into the 10s or 100s of giga Hertz - so low frequency signals (audio band in particular) are left to flail around and find whatever they can. Even a relatively low frequency fields (pure DC isn't affected) will have complex harmonics that can ruin your day, particularly where you have a high-impedance input such as the input pins of an OPA1642. Very high speed PCBs are often 6 or more very thin layers to create better capacitive coupling between the layers and contain those wretched fields were we can deal with them.

The maths behind all of this is way over my head but fortunately there are calculators to help. Although I use the term parasitic capacitor here, this one is useful. All components have unwanted parasitic resistance, capacitance and inductance and it's these reducing effects that dictate why those 100 nF local decoupling capacitors have to be physically close to the ICs.

Even things like the physical size of the component (SMD or THT) can effect the parastic effect so much so it's a wonder that some boards work at all! Rather ironically, there are times when THT components are useful so long as the leads are bent at right angles since the magnetic field that surrounds the wire is kept a couple of millimetres away from the board. With SMD you have to be a bit more careful.

Many simulators will allow you to put enter the parasitics and that can help catch these issues but there's nothing quite like making the board and "suck it and see".  

A very handy rule of thumb is to always use a four layer board even when the design only requires two layers. Then put a unbroken ground plane on the two  inner layers (layer 2 and 3) and route signals and power on the outer two. Very sensitive boards can swap that but that requires vias often place in the ICs pads and that creates wicking unless you pay for a board that supports it (typically 6+ layers from JLC etc.)

Without seeing your board layout I can't be sure if you've walked into any of these traps but if it doesn't work as expected, it's likely to be one or more of these.

@marcdraco 

Thank you so much for the detailed explanation, I really appreciate how you took the time to break things down even if some parts were still a bit over my head haha.

I chose the OPA1642 because I saw in the forum comments that it's considered a good replacement for what Jules originally used, and it's also easier to source here  which made things more convenient for me.

I'll be attaching the layout of my PCB here so you can see what I came up with (red lines are top layer connections while blue lines are the bottom layer connections). I still don't have a solid understanding of virtual versus true ground, which is probably why I got confused with the green arrows and ended up walking into that trap haha.

For the 1G ohm resistor, I’ll just buy one from LSCS, together with the OPA1642, since I saw they’re not that expensive in their site, and that way I won’t have to rely on a makeshift solution using graphite.

As for the zener, I did change its symbol to a regular diode in the schematic as it is the only symbol I saw that is close to it, so I’m thankful you pointed it out. I’ll be sure to use the proper zener when I start soldering everything, and I’ll try to be more disciplined when drawing schematics from now on.

I’ll also go ahead with your recommendation to use a 6V zener, since I’ll only be using the JLI2555 electret capsule for now. I’m not working with true condenser mics yet, maybe in the future when I can save up for them haha.

Again, thank you so much for all the guidance, it really helps beginners like me understand not just what to fix, but also why it matters.

We're all  beginners at some level, I made a joke about it before, but a journey of 1000 miles has to begin with a single step. 

You've probably seen the clip of Elon Musk boasting to my old boss (Chris Anderson, who's a lovely bloke and very smart) about knowing more about manufacturing than anyone else alive. Which is patently such "Dunning Kruger" that I almost face-palmed my head right off my shoulders. I felt a little nauseous watching that because Chris didn't challenge him and the audience just applauded him. Presumably someone, somewhere is the GOAT but it's not Elon... the man who promises everything but often delivers nothing.

As a science/engineering writer by profession, I know how difficult it is to grasp all this stuff even for experts. But I also believe that knowledge should be free for all who wish to learn. Within reason of course, everyone has to eat. But universities the world over are more like mills these days that take students in at one end and turn out people who know loads of theory but lack the know-how to put it into practise. There are odd ones like Dr. Eric Bogatin (University of Boulder) but he's the rarity. In science and engineering, we're constantly learning new stuff (if we're not, we're doing something wrong) and sometimes we have to throw away the old knowledge and replace it with better. Many experts are loathed to do that and we see old, outdated techniques appearing all over the place which tends to throw even the most experience-hardened experts into a panic.

As a brief segue, we used to think electricity passed through some mysterious substance called "ether" - no one knew what it was, they just figured we couldn't see it. They were right in that regard, we can't see it because (so far as we know to date) energy moves through a vacuum of space. Sounds bonkers, but feel the warmth of the sun on your skin on a clear day - that's electromagnetic radiation from the sun at a wavelength (energy level) that we perceive as heat. We can't feel higher energy radiation like ultra-violet but it causes our skin to get radiation burns - which is exactly what sunburn is. (Pale skinned folk take note - that sunburn you caught on your holibobs is a radiation burn...). Life on Earth would not be possible without the sun to power it but the very dangerous radiation (UV-C) is deflected around the Earth's magnetosphere. If it wasn't... 

And almost everything I've just discussed was discovered in the last century or so. Imagine being one of those early pioneers - like Einstein, Curie, Rutherford etc. and having to get the old guard to drop everything they thought they knew and replace it with something weirder than they considered imaginable!

OK, back to the meaty stuff.

When I started everything (except radio transmitters) was quite leisurely (low frequency) so it was pretty easy to make a PCB that worked first time, provided the schematic capture/design was right. These days things are so fast, switching currents measured in amps in billionths of a second, that the old ways simply don't work any more.

Although the V2 USB C mic only runs at 12 Mb/s the input lines have to be kept short and as close as possible to the same length. Miss that target and when the chip flips the polarity of the two lines (it's differential just like the analogue differential Matt and I have used on the mic inputs) one signal can arrive at the receiver and switch BACK before the other signal has time to get there. These signals are moving through the FR4 plastic at around 2/3rds light speed which means you have perhaps a couple of millimetres wiggle room. And as it gets faster the more accurate that match has to be!

Time for some theory!

As for your "mistake" regarding virtual ground, take it from me, that's on Jules (or whoever drew the original schematic). Best practice dictates that we use ground reference symbols for ground - and nothing else. Period. Using ground symbols is marvellous way to confuse yourself and others!

Just to confuse matters further there are multiple schematic symbols for ground reference too. Four come to mind:

Analogue ground (that's an open arrow) - sometimes marked GNDA.
Digital ground which is a like an upside down T shape, GND digital.

and less common (and even more confusing).

There's chassis ground (which is where the shield might connect if it's not a Faraday cage) this one has lines under a bar at 45 degree angles like a little bit of hatching.

Finally there's Earth - now this is TRUE Earth because it connects to your home's earth point and somewhere goes to a piece of copper that's buried deep in the ground near or even under your home. We never, ever use true Earth (ironically where the term "ground") comes from unless we're protecting against electrocution. Anything metal in your home such as a metal sink, the taps in your bathroom and so on are connected to this. In this way, should a stray live wire manage to hit the copper pipework, the fuse will blow. I had a "shocking experience like this decades back in an very old building when I went to do some washing in an older part of the home. The sink wasn't grounded and the instant I put my hands in the water... ZZZAP! Man that stung!

Here's a small selection of the available grounds (EU standard) available in KiCAD:

KiCAD calls "chassis ground" ground power (GNDPWR) here with the various other symbols for signals OR the true Earth.

Since you're a capable student (and for anyone else following), get used to calling "zero volt rail" or "ground reference" and it'll be much easier. The voltage tapped on a divider like this is a reference point and should never, ever be called ground.

The reason is that when we measure potential difference (voltage) the reading we get is referred (or with respect to) to zero volts on one side of the measurement. The simplest method is to measure the voltage on a battery pack (or even a little cell). We put the black lead on the negative pole and the red lead on the positive.

If we flip those over your meter will read minus voltage because there is less voltage on the red lead than the black.

Sounds simple enough, right.

One of the simplest components we use is a resistor and while they're more complex than basic theory would suggest, we know that when some current flows through a resistor, a voltage is "dropped" across it.

If you were to connect a resistor (any value over 1000 ohms is good. This way we don't pass enough current to make it hot. Now measure the voltage across the terminals of our battery and we get the same result - nothing particularly surprising there.

OK, now let's consider the setup we have with R8 and R9 on your schematic.

Note first that the two resistors are the same value. 

Thevenin's first law says that the sum of all the currents entering a point on the circuit is the same as all of the currents flowing out of it. (I used to call this Thevenin's law of the patently obvious because it was to me and it seemed an overly complicated way of explaining a circuit. The root word is "circle" hence a circuit is a loop that current just runs around.

The word current (named for Ampere) was used because electric current is flows just like water in a stream. A battery therefore can be thought of as a mountain with a sluice at the top and water flows down the mountain. The height of our mountain gives the water "potential energy" which you might recall from your high-school physics.

The water pooled at the top has 100% more potential energy than the water that pools at the bottom. In nature the energy of the sun causes water to evaporate and collect as clouds, so the water in the clouds is an energy store! It might take a few goes to wrap your head around this (took me ages) but once the penny drops it just seems, well, obvious.

Staying with our mountain stream, as the current flows down the mountain, the water loses potential energy swaps it for kinetic energy - the speed that the water is flowing and when it hits the pool at the end, all of that energy has gone.

So the energy moves from the top (positive) to ground at the bottom. In electronics we call this the potential difference or voltage (named for Voltaire). The higher our mountain, the greater the potential difference so the greater the speed of the water running down - more current.

Now imagine that water was flowing into a huge bathtub and then someone pulls the plug. The water flows out of the plug hole gathering speed, but relative to the bath it's flowing out - hence the potential difference of the outflow is negative.

The currents are the same but the voltage is determined with respect to the reference. This would be better explained in a video but I don't have the skills to make one in double-quick time, so we're stuck with prose.

Now what about R8 and R9?

Since the bottom of R8 is connected to the circuit ground reference and the top of R9 is connected to Vcc (positive supply) electric current flows through from the top of R9 out the bottom and continues on through R8 until it connects to the bottom of the circuit and carries on looping around.  

Here's the circuit in the Falstad simulator (which often gets a bad rap, but it's darn handy when you're starting out). The link will take you to the circuit in Falstad where you can see an animation of the current flowing. In this simple simulation current doesn't flow through the voltmeter and no voltage appears across the ammeters because (a) that would make it overly complicated and (b) see a. 🙂 Reason being these voltages and currents are so incredibly small (on modern gear) that they don't affect the circuit (this is called "loading").

https://tinyurl.com/2xq4t686

You can see here that around 50 micro amps is flowing in this circuit - the voltage source is a "perfect" five volt source. In reality, voltage sources have an internal resistance - another annoying bit of theory that really means that a power supply can only supply a limited amount of current before its voltage droops to the point when the circuit won't function (and the supply burns out, literally in some cases). Terms like "burning out" or "brown outs" literally come from times when the early experimenters (and people who forget about this) find their parts going up in smoke and plastic leads going brown due to burning!

This is modelled in more advanced simulation as a resistor that you can't see hidden in the voltage source. For now, we can dispense with such detail and in a properly designed circuit it shouldn't be a problem.   

OK, so we've got 50 micro-amps flowing through both resistors. So we can use Ohm's law to calculate the voltage "drop" across each resistor. I've put a virtual voltmeter on here so you can see it measuring 2.5 V at the junction of R8 and R9.

Ohm's law (guess who that's named for) is a magic triangle where given two values we can calculate the third through simple division or multiplication. For voltage that's the resistance in Ohms multiplied by the current.

So we have:

47,000 ohms * 0.00005 amperes (grabs calculator):

47 x 10e^3 * 50 x 10e-6 (which isn't well formatted but your calculator or a simple spreadsheet will let you use exponents with the E button.

Or 2.35 volts. The error is because I rounded the current, but you can see the voltage at the junction (or across each resistor) is equal to 2.5 volts. This is one of the simplest circuits we have and it's called a voltage (or potential) divide because it divides the voltage into smaller parts.

If we were to add more resistors or change the value of these the voltage at the junctions changes in turn - but the current in the circuit is always the same at the entry and exit (in Falstad):

https://tinyurl.com/26v4tn8a

If you add up the total resistance in each example you'll find it's the same and even though there's bunch of different values, the current flow is exactly the same and the voltage at the Virtual Ground node is still the same.

For the next example, I've put volt meters across several of the resistors so you can see how the voltage "dropped" across every resistor of of the same value is the same voltage but the current is constant around the loop.

https://tinyurl.com/25ztt2l7

In each case there the voltage is measured across the resistor - black lead to one end and the red to the other. If you know the value of a given resistor and you measure the voltage dropped across it, you can measure the current flowing through it. Pretty wacky eh? This is part of basic fault-finding when you have something that doesn't work, your own design of even something you've never seen before.

As an aside your digital meter measures voltages (old, analogue meters only registered current so they were dead easy to burn out if you accidentally put too much through the device and the fuse didn't go - which would often be the case). This happened to the first multimeter that I made and the meter movement was too expensive to replace so it ended up in the trash. The value you receive, be that voltage, current or even resistance is calculated using Ohm's law by the computer which contains a very accurate ADC.

TL;DR

All of this wordy explanation is to prove that "ground" (zero volts) is imaginary. It's some point on a graph that we say is the starting point. So anything above that point is positive and anything below is negative.

If we call our most negative point zero volts then all the other measurements change accordingly and this is why it's called a virtual ground or ground reference.

So why do we even need that? 

It's because operational amplifiers are bipolar internally. Their idea of "ground" or 0V is the an imaginary centre point half way between the two supply voltage rails. The IC's internal circuitry expects current to be able to flow in either direction and this is why we need to have a "negative" rail.

But if we don't (as here) there's only a positive supply - say 6V - the "total" supply voltage. Some op amps are designed to be able to run with a single supply but for now let's stick to the black box which has two inputs, one output a positive supply input and a "negative" supply. 

Note the two voltmeters near the ground reference node. The negative end (black on your meter) is connected to the ground point, so one measures +5V and the other measures -5V. No trickery, the total voltage (the two voltage sources in series) give a total voltage of 10V but we've moved the measurement point to the join the centre, so current (each voltage source is independent) still flows around the loop in the same way but if we measure reference to that middle point we have split 5 and -5V rails.

Magic!

The operational amplifier (not shown) can now deal with A/C current which swings about the reference point - and this is what the virtual ground does using a bit of trickery caused by how these amazing devices work.

You'll note that there are two inputs on the triangle - + and -. Which means "non inverting input" and "inverting input" respectively.

In simple terms any voltage (op amps are, for the most part, voltage operated devices) you apply to the non-inverting input appears at the output multiplied by the gain of the amplifier. (In the Alice both devices have a gain of +1 and -1 respectively). They're called "operational amplifiers" because they formed the basis of analogue computers - and anyone who thinks, "how quaint", for some applications analogue computers are faster than the fastest computers on the planet with the exception of quantum tech, those computations happen at almost the speed that the currents travel through the wires.

If we were to apply 3V to the + input, the output would also read (shocker) 3V.

Now imagine we put 3V on the - input? In a dual supply system the output would flips and -3V appears.

And this is where that reference comes in. Somewhere in the circuit, there's a ground reference that the amplifier.

If our ground reference sits in the middle of the circuit. Iit's convention to draw a schematic with the most positive supply at the top and the most negative or zero point at the bottom. This represents conventional current flow with electrons flowing from the positive electrode to the negative one.

But imagine if you will the ground reference sits at 0V - with a total supply of 6V and we apply 1V to the inverting input? Then what happens?

As clever as op amps are, they're unable to break the laws of physics, so the output will go (pretty close) to 0V and just sit there.

Which is clearly the wrong answer - and (finally) the point of the virtual ground comes into play.

When we apply (by convention) half supply voltage to the non-inverting input the amplifier still sees 1V at its inverting input now it can pull the output down below that to the "real" 0V rail, giving us a -1V answer.

This is called biasing the amplifier and it's something we take for granted in op amp circuits although it's something that gets rather more involved if we're using transistors. (I won't go into that because it opens a whole factor of worms and, to mix my metaphors, we don't want to go down that rabbit hole!

So in conclusion, the virtual ground adds a voltage to the non-inverting input and pushes the internal reference up, allowing the amp to perform it's magic. 

Now you may be thinking, 3-1 = 2 but that's not what the amp "sees". So far as the amp is concerned, the 3V is half the supply voltage and that becomes the 0V reference for the calculation. It's a bit weird at first but trust me on this.

All this extra complexity is why most engineers avoid using single supply designs in A/C circuits. It's far and away easier to produce the reference ground at the power supply with the two power rails sitting at + and - with respect to that. There are a lot of man traps waiting for the unwary in single-supply design.

Speaking of ground references, the THAT151x devices are also operational amplifiers but note that Pin 5 is called the Ref pin and (shocker) that's connected to ground in Matt's design.

Although it's unusual, there are times when you need to alter the reference ground with respect to the OUTPUT while keeping the INPUT ground at the same potential (usually within a few volts) and that's what that pin does. Clever fellas these.  

Excellent idea - I assume you're not having the board assembled at JLC?

Posted by: @renzevan

↑

As for the zener, I did change its symbol to a regular diode in the schematic as it is the only symbol I saw that is close to it, so I’m thankful you pointed it out. I’ll be sure to use the proper zener when I start soldering everything, and I’ll try to be more disciplined when drawing schematics from now on.

The nearest actual Zener you'll get is probably a 5V6 which should be fine. I think the next one (from memory) is 6V2. If you have the cash (and you have loads of board space) an LDO is a better bet. Line regulation is far better and they (generally) produce less noise than a Zener. You can get fixed 3 terminal LDOs with various common voltages like 5, 3v3 and even 1v8 (for really, really low power stuff). 

There's also the TL431  (or CJ432 at LCSC - the pinout is different on the CJ431 - and yup, I walked right into that lamppost myself). It's available in TO92 and SOT-23 SMD versions and only requires a couple of resistors to set the voltage. It's a shunt regulator (like a Zener) and sets the output to whatever is required to keep the reference pin (fed by a voltage divder) at 2v5.

So, for example, taking a 10-15V supply (where you don't know exactly what you're getting) and you want 5V out, you just use equal value resistors - just as in the virtual ground as discussed above. Note that the "impedance" at this pin is far and away too great to use it as a virtual ground which is a shame.

As with a Zener diode, a shunt regulator operates by drawing enough current from the supply so that it buckles (see earlier discussion). Now clearly we don't want to short circuit the supply or something is going to break!

As with a Zener (which becomes almost short circuit at it's set point) we need something to stop that happening and that's a resistor.

This is probably the most difficult part of the calculation (I've tripped up over this by being lazy, guestimating and getting the value far to low and lo-and-behold, the magic smoke appears ).

A couple of things need to be calculated.

First of all we need to know the to maximum current the circuit we're supplying is going to draw - note, not the quiescent point (that's the current used when the thing is sitting there not doing anything). For the OPA1642 that's a max of 2.3 millamps per channel so a dual like this is going to draw a max of around 5 mA (always round up where you can to give yourself room for manoeuvre).

During operation the outputs appear to have to drive 47R resistors into the load which sounds barmy - that current is coming from the supply too - but this is a little misleading as the supply is fed (at the phantom end) by two 6K8 resistors in parallel for a maximum current of 14 mA (into your circuit) although it'll never get there because if you try to draw 14 mA (and you can't because there's an additional pair of 2K2 resistors in there) voltage would drop to 0.

For practical purposes, P48 phantom mics need to draw no more than 5mA which is more than the quiescent point of a single 5532 so we have to use low-power amps (or transistors which are more efficient if designed properly but less predictable for reasons beyond this discussion).

So let's say that we have 5 mA for our circuit, what's the actual input voltage when those resistors are all connected.

Each line has 6K8 plus 2K2 = 9 K and we have two in parallel to that's around 10mA - that's IT and remember if we try to draw 10mA t

48V across 4k5 = is roughly 10 mA, and we need 5 mA.

So a 5mA load means we have a "loaded" voltage of 24V (approximately).

This is where the magic happens:

Let's say we're using a TL431 and we've set the reference voltage to 2v5 so the output is 5V. 

Now subtract 5V  from the voltage we have at the input after all those resistors.

24 - 5 = 19V

Rather than yanking on the supply and hoping for the best (which is bad design and praying for a good result 😳, [guilty your honour]) we need a dropping resistor to get shot of that 19V while still allowing enough current to get to our circuit.

So we take 19 V  that we need to lose and allow it to pass 5 mA which is a (*finally) simple matter of using ohm's law:

19 V / 5 mA = 3800 ohms.

The nearest preferred values are 3k6 and 3k9 (on the E24 scale, more precise ones are available but this is plenty accurate for our needs).

I'd use 3k6 or even 3k3 (a popular choice) in this example since the circuit is only drawing 5 mA which gives us a little bit of wiggle room should the circuit require a little more that we've estimated from the datasheets. (Advanced simulators like LTSpice can help here.)

Any excess current is passed through the TL431 and lost as heat but it's such a small amount even a 1/8 W resistor wouldn't keep a flea warm.

While simple, shunt regulators are not very efficient (at least they are cheap!) LDOs, particularly the fixed ones can operate within a few millivolts of the supply. Just be aware that the more popular ones often draw significant quiescent current and that's something you don't have to spare.

OK, now the theory is out of the way, let's take a critical look at your board and see where it can be improved. I've made every mistake here (and loads more) so don't worry it's how we learn!

That board will fail DFM (Design for Manufacturing) check before it even got to the factory floor. So if you haven't ordered it yet, please don't! DFM is automatic and included on any design package worth using - including EasyEDA and KiCAD.

The main reason is you've mounted the resistors vertically - never do this unless you absolutely have to and even then, see if there's somewhere else they can be placed. 

But worse than that the rings around the components show you a bird's eye view of where they will fit - anything that clashes here is going to clash on the board and you'll find you won't be able to get them without tying yourself in knots!

The other reason why it's bad practice to mount them vertically is that it's dead easy snap them off, leaving you with a non-functional board and a little bit of wire stuck in the hole! Where possible, always mount them horizontally, flat on the board so this doesn't happen. Ideally (and there are tools to make this easier, from 3D printed ones to special pliers, bend your leads at right angles to the board so the part drops in. On 2+ layer boards you can pop some solder on the top layer and that will hold the part in place while you flip the board over and do the reverse.

SMD parts are far easier in this regard but the little beasts are often small and very difficult to handle. Although parts exist in 2012 and larger, you'll often find the value you want isn't made in that size so you're stuck with 0603 or heaven forfend, 0402! (Smaller ones exist but you need a precision board for those - and no one in their right mind would try that with a soldering iron!

I salute you using radial leaded (upright) capacitors as they use up less space, so that's a winner. Again, check with LCSC that you have the correct footprint for the voltage and value you've specified. I've just re-done the Katie (my version of Matt's design) with the smallest (!) 6,800 uF low voltage capacitor I could find and it's got a footprint diameter of 18 mm and don't get me started on the height of that thing! 

Getting into the nitty-gritty stuff now, there are a couple of other man-traps you might have walked into. 

The 100 nF decoupling cap that supplies "local" charge to the IC should be mounted as close to the power pins as possible. Looks like you've picked a solid polymer cap here. They're quite expensive in real terms and gobble up board space. 

Here's a section from my latest WIP:

The larger part is an (experimental part so ignore that) but see how the capacitor is mounted within a gnat's willy of the IC's power pin. Power distribution is the blue (outer Layer 4) and signals are mostly outer Layer 1. Layers 3 and 4 are dedicated to 0V reference, no actual signals go anywhere other than through there.

The purple outlines in KiCAD are the courtyards - any component has to be placed so they don't overlap. I'm pretty tight on this board so I might get a "NO!" from JLC's DFM. When they said "as close as possible", I might have taken that a little bit too literally.

Here's the 3D preview which is dead handy as you can see where the actual components fit:

So that's about as close as the Pick and Place machines will do it (and right on the limit for "Economic" boards).

There's another gotcha! Here near the OPA1642 where you've changed sides with a via.

Vias holes drilled in the board and then plated with a thin layer of copper. As such you need to keep them a little bit away from the parts wherever is practical. The drilling machines are amazing but they do can skid slightly off centre (they round the position up or down to whatever tolerance their are capable of so your vias might not (at some level of magnification end up where you thought they are!

This can cause a few maladies, some obvious, some less so. Keeping your vias a little distance from the IC pins is a good idea but the one on Pin 6 is a bit too close to the fan out of Pin 7.

There are two issues here - the "obvious" one is if the drill was off by a hair you'd short 6 and 7 ...

Where high-speed (or sensitive) boards are concerned there's a demon waiting around the next corner. Here it is as a close up of a 48 pin SMD chip with 0.3mm wide pads and a lot of high-speed signals (this advice applies to any board, but mostly to high speed stuff).

 
On the face of it, it looks like a mess of wires with vias clumped all over the place but when you look at the X-ray view with the void space around the vias the problem comes into view.

I almost fell right into this hole myself (I caught it doing my revision using a talk from the wonderful Phil Salmony).

This is one of those "huh?!" moments that only makes sense when you think in terms of fields (electric and magnetic). The electric field is the one we're interested here because it needs to return via the copper immediately below it. (See the earlier discussion on high-frequency stuff.) At audio frequencies this is rarely going to be a problem but as I found out in the School of Hard Knocks, University of Google... when you're working with digital stuff that little beastie jumps right out.

To keep those fields in check (because if we don't coral them, they'll find a a way out themselves and interfere with other signals!) And to do that we need to have solid as much copper directly under the tracks AND around the vias.

Pin 48 (a control pin) has to be pushed around the via on pin 47 (a high-speed data line) because if it that track crosses a void space there's a chance the field could jump onto Pin 47 and cause bit errors. It's unlikely, but it can happen.

Current best practice for very high speed designs dictates that any signal via that has to transition layers should have a grounding via placed as close as possible to it to contain the fields that can "leak" from the signal via. One is missing here... can you spot it?

Here's the  correct" way to do it:

These vias are quite large in real terms but it allows me to produce a more cost effective board without worrying about making the hole so small it's at the very limit of what the board house can produce.

And alternative is like this:

That single via can help to contain the return currents from each layer transition and allow them to flow safely to ground. However, I suspect while this passes DFM, there's a slim chance the plating will leak down into the void space (see how the drilled hole is almost clipping the void)? If that drill is off by a tiny fraction (about two 100ths of mm) it will cut the void space on layers 2 and 3 and short one or both of them. Result? "Me board, he not worky!"

And that's always assuming I have the thing wired correctly! I've made more "whoopsies" than I've had hot dinners (and judging by my dad bod, that's a lot of food)

If you look at the board side on - it's easier to imagine the fields spreading out. Ideally we need to trap them on the next available layer in the stack Four layer boards are cheaper than ever these days so it's false economy to use anything else. Route all of your power and signals on the outer two and keep the inners for ground only. I'm re-doing some of my earlier work to follow these guidelines myself, now I have a better grasp of the problem. (Again, thanks to the likes of Rick and Eric.)
 
I'm going to ask Rick Hartley this when he gets back and we'll get a better answer from someone who has literally designed thousands of working boards - many of which are used in commercial aviation where they can't go wrong at any price. 

All in all, and my comments aside, I think you've done a smashing job for someone who's new to this and fingers crossed, it will give you years of excellent service.

While I was doing the updates, I did a quick 3D rendering of Katie (four layer) to show you some of these "best practices" in action - see how all the resistors are laid flat on the board. Signal and power are routed top and bottom and there's a clear ground plane on layers two and three as you can see in the X-ray view.

Posted by: @renzevan

↑

Luckily, all my measurements were correct! Although the fit is tight, it still fits HAHA. I just soldered it without pushing the PCB too close to the back of the capsule to avoid blocking any holes on the rear part of the capsule.

Sorry, I forgot to say, that's a fine job there young man! I'm glad we have something like these for the new design because it makes it far more accessible.

If the bug has you (and it seems to have) you're going to have to figure out some way to explain things to your girlfriend. Congrats on your first boards but I just know this is the first of many. 

As for the money (as I've told my family and even Matt) it's a drop in the ocean compared to what you'd have to pay to sit in a classroom all day. Sure, you won't get a qualification this way but if you keep it up, you'll probably find yourself a job just based on your skill. 

Being handy with a soldering iron and the rest of the kit you'll assemble as part of the hobby, you'll soon find yourself the local go-to guy for getting people's electronic repairs done. Some folks go full time, but most offset the cost of better equipment to make their hobby easier and more enjoyable, so it's a win-win!