ESP32-S3 design review - No TX

[Thalhammer]

Hi,
Please be gentle, I have never done RF design before, not have I studied EE.
I also know that ESP modules exist that include the antenna and crystal, however I wanna push my boundaries and make my pcb small/clean.

I designed a PCB around the ESP32-S3 together with a IMU and a compass chip as well as a USB-C connector so that it can be directly plugged into a device like a phone and get power from it.
I obviously made a bunch of stupid mistakes in the first revision (like forgetting to connect one of the power pins) and I am going to roll a second revision of the pcb where I also optimized a bunch of stuff along with fixing the issues.
For now I have corrected them with solder bridges and bodge wires in the right places and I am at the point where I get the main ESP32 chip up and running, as well as connected to USB and the ESP32 can see the I2C devices (I havent tried reading data from them yet, but since a i2c bus scan shows them they are probably fine).

My problem is as follows:
I can only receive data but not send any.
If I do a scan for all the nearby wifi networks I get a full list of all the ones that send active beacons and the indicated RSSI is actually pretty good.
However if I try to connect to them or send any 802.11 frames (using esp-now, without a connection) I can not see any packets using a wifi card in monitor mode.
Trying to connect to an AccessPoint fails with an auth timeout, indicating that the AP never got the request for the new device.
Flashing the identical code to a random premade ESP32-S3 board in my drawer proves that both programs (and my testing methods) work perfectly fine and that indeed the board is simply not emitting any RF.
Since I did the RF part by basically copying other peoples schematics and PCB's I already expected it to perform poorly, but I kinda expected it to at least emit anything to a device right next to it on the same desk.

Below is the schematic and layout of the (very simple) RF part of the chip.

Sadly I don't have a oscilloscope or RF analyser so I can't really tell if theres any emissions of the antenna. I can only judge based on whether my notebook receives any frames.
I do have a salea logic but its nowhere near fast enough to capture a 2.4GHz waveform.

Any obviously wrong things and further steps to debug it ?
I did verify that none of the antenna pins are shorted to any other pins.

Thanks in advance,
Thalhammer

[PlainName]

Might not be your entire problem, but the tracking between the parts (ESP to L3) doesn't look like it's a stripline. The combination of the track and ground plane (and surrounding ground) should give a 50 ohm impedance, and since yours lacks the ground plane...

In practice, having the track more or less the same width as the components is useful since that reduces reflections. You can see from the AN3216 datasheet that the illustrated layouts on page 8 use very fat tracks (and stitched ground plane).

[PlainName]

so "L4" is the 2.4GHz ceramic resonator?

I think you mean L3. And, yes, it needs ground termination 

[Thalhammer]

So how do I fix that ?
Add a ground plane below the tracks ?
Should it be a full ground plane or should it have a gap to the main one (like for the crystal)?

> should give a 50 ohm impedance, and since yours lacks the ground plane...

I have heard about that before, but how do I calculate that ?
Like I said, I am new to RF, the fastest I did so far was USB full speed and at distances short enough to yolo it.
I always assumed you are not supposed to have a ground plane below or next to RF things so thats what I did.

While we are at it, what are good spacings between the chips LNA output and the antenna ?
Is there a certain minimum I should keep (to avoid interference) or can I place it as close as physical possible (while still being able to solder it)?

Are there any good sources for information about things like that which are understandable without a degree in EE ?
If my antenna design is so wrong, why can I still receive just fine ?

How can I measure/verify if my design matches the requirements ?

Thanks so far.

[PlainName]

Your PCB design package may include an appropriate tool, but if not you can use an online calculator. There are many, of which this is the first I found:

https://www.allaboutcircuits.com/tools/microstrip-wavelength-calculator/

For minimal ground plane and component spacing info, the datasheet would normally give some details. I note that many of the accessible datasheets around don't, but the one on LCSC (attached) does. You need to have at least the ground plane shown and, importantly, not put anything in the clear areas.

Without appropriate test tools I think the way to find out if you've got it right or not is whether it works. Receive can work with really very bad antenna - it has to work on very low level and crappy signals. Transmit, on the other hand, has to be a good signal so it can still be received far off in the midst of noise as that crappy signal.

[ftg]

How is the power integrity when attempting to transmit?
Does it pull the expected amount of current and how big is the current spike.
Some low cost sdr like a chinese hackRF clone (called Cirket or Clifford now on aliexpress) would show you if anything gets emitted.
Because even a bad antenna should radiate something.
But if there is a problem with the power integrity, the packet itself might be malformed and thus not decode at the wifi ap.

[Kean]

Chip antennas are very reliant on correct PCB and ground plane layout, which is why they can be so small.  They are not always going to work sufficiently well "out of the box" on a new PCB layout without tuning using a VNA.  You should study the datasheet for both the ESP32-S3 and the chip antenna to review impedance matching and layout recommendations and try to implement as many of those recommendations as possible.  Even better is to follow their reference design which will be know to work reasonably well.

With transmission, a bad impedance match can lead to excessive reflections during transmission.  In some cases this could even lead to damage of the RF Tx part of the IC, although that may not be the case here and it is just performing really badly.

[Thalhammer]

> Even better is to follow their reference design which will be know to work reasonably well.

I would, but unfortunately all the reference designs from espressif that I could find use pcb antennas or a ufl connector both of which I'd prefer to avoid to keep the pcb small.
I have revised my layout and switched to a 4 layer board (top and bottom for signal, inner1 is a GND plane, inner2 is a VCC plane). Which allowed for a cleaner and most importantly a proper ground plane for the rf strip.
I have used jlcpcb's impedance calculator with the stackup I am using and calculated a trace width of 13.75 mil which is the size I am using now. I also rearranged the rf strip to prevent sharp corners.
I also made sure the spacing between the ground planes and the antenna is at least 3mm as noted in the antenna datasheet.
See below for the current layout. I also improved the power layout (even though I didn't see any significant ripple with my salea) with thicker traces and moving the caps to the top layer where possible.
I'll probably also add a bigger cap if I find space somewhere just in case. While I was at it I also added a bunch of test points for debugging as well as length matching the usb pair (even though USB worked fine).

I am hopeful that this pcb will work better (but so I was with the first revision, I really have no real clue if I am doing this right), but before I order it again I'd prefer if you guys could do another review of it.

Another thing I am extremely unsure about is the following:
What should the matching network look like ?
Currently I use a inductor and cap in series (see first post), however a lot of designs I found (in fact the majority) use a single inductor with a cap on either side against GND.
This is also what the ESP datasheet and the antenna datasheet recommend.
Whats the difference ?
How do I have to size the components for it ?

[Gribo]

A Pi network (2 shunt capacitors and a series inductor) will allow you easier impedance matching. Worst case, don't populate the capacitors or short the inductor.

[Thalhammer]

So should I change it to a PI network ?

What about the rest of the changes ?

[iMo]

For example the cheapo C3 boards with the smd resonator antenna are constantly criticized by the makers.
I've seen vids they replace them with piece of wire creating the matched antenna to get a reliable signal over couple of meters.
I would go with standard "meander like" pcb antenna when possible (MIFA?)..

[floobydust]

There's a lot of science, engineering and testing to do an RF antenna design and troubleshoot. I'm not an expert but would say you won't learn it from a bit of text in a forum. That leaves a bit of research and copy existing designs.

You could move your board close to the WiFi source and see what you get, if it's just weak or there is zero TX signal. Or solder a wire stub like the mod below and see what happens.

ESP32-C3 uses CrossAir CA-C03 chip antenna. "... modification involves adding a 31 mm length of 1.0 mm silver-plated wire, configured as a quarter-wavelength (λ/4) antenna."
https://peterneufeld.wordpress.com/2025/03/04/esp32-c3-supermini-antenna-modification also in Elektor  magazine.
https://docs.espressif.com/projects/esp-hardware-design-guidelines/en/latest/esp32c3/pcb-layout-design.html?utm_source=perplexity#rf

Keep in mind your TX problem may not be due to the antenna or feed. I don't see the schematic or rest of the PCB pics for what you have made.
Espressif gives many factors affecting the TX performance such as crystal type and PCB layout for it (TX cause freq. shift).
3.3V rail noise (but I find it seems to just crash).
More here: https://docs.espressif.com/projects/esp-hardware-design-guidelines/en/latest/esp32/pcb-layout-design.html
Note the ESP32 requires a CLC filter on the output, not just a PCB trace but a stub too.

[Thalhammer]

You could move your board close to the WiFi source and see what you get, if it's just weak or there is zero TX signal.

I tried that. I loaded a sample that sends esp-now packets (1Mbit/s, fixed channel, 20Mhz bandwidth) and tried to capture them with my notebook wifi put into monitoring mode. If I use the sketch on a different dev board I get the packets, if I use it on mine I get nothing. So I'd say there is zero (or close to zero) data emitted (or if it does then the frequency is way off).

Or solder a wire stub like the mod below and see what happens.

I'll try, but I suspect it won't help much, because if it was just weak it should get received at least from the notebook next to it.

Keep in mind your TX problem may not be due to the antenna or feed. I don't see the schematic or rest of the PCB pics for what you have made.

I have attached the current version of the PCB Layout and Schematic below, the old version had a bunch of oversights which might or might not be responsible for the issues, but since I fixed those in the schematic and layout I assume it makes more sense to post what I intend to send to jlcpcb next instead of the broken one.

Note the ESP32 requires a CLC filter on the output, not just a PCB trace but a stub too.

This is good to know, the more I browse the web, the more I suspect that the schematic I "got inspiration from" was never used in the real world, because the setup they used for the antenna trace and a bunch of other things clearly seems wrong.

If anyone wants to read the full schematic below I'd really appreciate it (I'll also take notes that are not related to the RF setup) but I don't wanna waste anyone's time with my stupid beginner mistakes.

[nctnico]

I have a CE/FCC certified ESP32 design with a ceramic antenna under my belt so I have some hands-on experience I can share:

There are a few things to do to make such a design succesful:

1) Understand RF PCB layout. Traces between the ESP32 and the antenna need to have a 50 Ohm impedance. Using an online PCB trace calculator is good enough for a standard 4 layer stackup.
2) Follow the PCB layout guidelines of the ceramic antenna to the letter (or at least as much as physically possible) and assume the ESP32 LNA pin has a 50 Ohm impedance.
3) An RF harmonics filter will be needed. The CLC circuit shown in the ESP32 datasheet is not primarily for antenna tuning but to filter harmonics. An easy way out is to use a readily made harmonics filter.
4) Get a spectrum analyser to get an idea whether you get some RF output. Even if it is just a cheap NanoVNA. A spectrum analyser capable of working up to 15GHz is preferred though as this can catch most of the relevant harmonics.

[floobydust]

Where is the ESP32-S3FN8 epad pin 57 grounded? I see 6 via's but nothing connected to GND or the red layer ground pour. I can't figure out the PCB pics.
I don't see any external FLASH memory IC connected. ESP32-S3 datasheet

[Thalhammer]

Where is the ESP32-S3FN8 epad pin 57 grounded? I see 6 via's but nothing connected to GND or the red layer ground pour. I can't figure out the PCB pics.
I don't see any external FLASH memory IC connected. ESP32-S3 datasheet

The pad is grounded using the 9 vias in the middle that connect to the ground planes on layer 2 and bottom.You are right that somehow the export doesn't show the vias, I'll recheck if I can somehow export an pdf that contains them in easyeda.

You are right that theres no external flash connected to the esp. Thats because the esp32-s3fn8 contains 8MB of flash on die which is plenty for my application.

I have a CE/FCC certified ESP32 design with a ceramic antenna under my belt so I have some hands-on experience I can share:

There are a few things to do to make such a design succesful:

1) Understand RF PCB layout. Traces between the ESP32 and the antenna need to have a 50 Ohm impedance. Using an online PCB trace calculator is good enough for a standard 4 layer stackup.
2) Follow the PCB layout guidelines of the ceramic antenna to the letter (or at least as much as physically possible) and assume the ESP32 LNA pin has a 50 Ohm impedance.
3) An RF harmonics filter will be needed. The CLC circuit shown in the ESP32 datasheet is not primarily for antenna tuning but to filter harmonics. An easy way out is to use a readily made harmonics filter.
4) Get a spectrum analyser to get an idea whether you get some RF output. Even if it is just a cheap NanoVNA. A spectrum analyser capable of working up to 15GHz is preferred though as this can catch most of the relevant harmonics.

1) I think the new PCB should be matched to 50Ohm impedance. I calculated the track width using jlcpcb's online calculator using their 4layer stackup.
2) I did my best to follow it in the new design, but I am very much unsure about things like that. I feel somewhat confident for normal stuff but everything HF/RF somehow feels like pure magic.
3) Ok so since I don't have anything of the likes of a CLC circuit it probably makes sense that it sucks. You mentioned that there are premade ones, do you have a recommendation ? Since I handassemble the boards having a single part should make it easier to prevent issues due to soldering.
4) I have considered getting a NanoVNA or some SDR that can do wifi frequencies, but even if I get one I have honestly no idea what I have to measure to figure out how to make it work.

[nctnico]

The harmonics filter isn't needed for the ESP32 to work so don't worry about that if you don't need certification. I used the DEA162450BT-1295A1 filter from TDK.

The best way to test with a spectrum analyser is using a direct connection. So remove the antenna and solder an SMA connector to board. If you see some signals in the BLE band when the ESP32 is on, the ESP32 is transmitting.

[Kean]

Typically you would work with application engineers from the manufacturers of the RF chip (Espressif) and chip antenna (RainSun).  Unfortunately unless you are going to be building these in large quantity (10's to 100's of thousands), those companies will probably have little interest in helping you.  In addition to the harmonics filtering mentioned, the chip antennas often need specific impedance matching to perform well at the required frequency because they are not even close to a quarter wave design.

Look at the RainSun chip antenna datasheet and check out the recommended and acceptable layouts.  If you can increase your PCB size it might be worth it, because it currently may be too small to work well.  The datasheet also shows some typical measurements of the antenna with a VNA, which you could try repeating with your PCB layout to see how well it performs - but that is quite an advanced topic if you are new to RF.  There are a bunch of videos on Youtube and discussions here about measuring antennas with a VNA.

That said, if you can make the PCB larger, then I'd suggest considering moving to a pre-certified module with PCB antenna.  You then don't need to worry much about RF design, which is quite a big field.  It also helps significantly if you ever want to move into production and sales of the design.  Your current PCB is almost certainly going to fail, and the cost of testing a design using bare IC rather than a pre-certified module is significant.

I tend to stick to RF modules from Espressif (e.g. ESP32-S3-MINI-1-N8) and u-blox (BLE/GNSS), and I have had quite a few designs all go through certification with only minor issues.  My typical customers generally don't want to pay an additional $20k+ on RF certification on a first generation product, but might consider it on a second gen design if there is proven market and decent cost savings per unit.

[ftg]

Before getting into antenna matching and measuring it, I would look at the supply rails during wifi package transmission.
And if there is no significant problems there, then consider getting a TinySA Ultra or a HackRF clone to actually see if anything is transmitted and how the transmitted signal looks like.

If a wifi dongle can have a working two way connection with a bare u.fl connector to a bottom of the barrel wifi ap across multiple meters, the laptop should see at least some of the packages sent by the ESP32 that is right next to it.

ESP32's are known to pull a decent amount of current from their supply when transmitting a wifi packet, so a multimeter is not enough, it needs an oscilloscope.
The good news is that it does not need to be super fast, just digital with a usable trigger.
As we just want to see how the supply rail behaves when sending a wifi packet.

If you need a very cheap way to measure the relative RF output power, a simple diode detector will work.
For example:
http://www.dl5neg.de/diodesensor/diodesensor.html
https://www.infineon.com/assets/row/public/documents/24/42/infineon-an-1807-pl32-1808-132434-rf-and-microwave-power-detection-applicationnotes-en.pdf?fileId=5546d46265f064ff0166440727be1055

If you also buy some SMA pigtails for soldering them in and some RP-SMA to SMA adapter, you could compare some other wifi device, or esp32 module with u.fl connection to your board.

But to get the most out of the diode detector, a digital oscilloscope is kainda required, to see the power envelope across the whole wifi packet.
A multimeter in peak hold/max hold might not react fast enough to the short transmission.

[floobydust]

At some point I have to wonder if the firmware is properly setting up the transmitter? There are many registers.
There is also the ESP RF Test Tool software for compliance and signalling tests that may be useful.

[Thalhammer]

At some point I have to wonder if the firmware is properly setting up the transmitter? There are many registers.

I am not doing anything special and the same code is working fine on another board with the same IC.

Before getting into antenna matching and measuring it, I would look at the supply rails during wifi package transmission.

I checked that with my logic analyser and couldn't see anything bad. I also tried turning the tx power all the way down to 10dbm and it did not change anything.

That said, if you can make the PCB larger, then I'd suggest considering moving to a pre-certified module with PCB antenna.

I know that using a premade module is way easier, but I kinda wanna learn how to integrate the soc directly. Plopping a module in is trivial. I am not buying the size argument either, because I have elrs receivers that are based on esp chips (esp8265, but I doubt the rf part is that different) and they also manage to have usable wifi. Again I am not aiming to connect over 10's of meters, the goal is literally to connect to another device <1m away. I also found https://www.waveshare.com/wiki/ESP32-S3-Zero, which is even more crammed then my pcb and still seems to generate usable wifi.

At this point I am wondering if I might have simply killed the transmitter when bodging the missing supply tracks onto it.

Since I have already fixed the vcc issues (and a couple others) in my current design version I'll just add a clc network instead of my cap+inductor pair because all the chip antenna designs I can find online that actually work seem to use a clc one. I'll reorder the new pcb and reassemble one using new components (I still got 4 more soc's in my drawer). Maybe it works better then, if not I'll be back asking for help ^^ Or probably just embed a esp32-s3-mini-1 which is about the size of my PCB.

[Randy222]

1st thing, make sure the code uses lowest Tx setting.
The default power can fry/swamp the LNA if the VSWR is bad. This could be the issue.

What ceramic antenna is it?

[Thalhammer]

What ceramic antenna is it?

Rainsun AN3216-245 Datasheet: https://wmsc.lcsc.com/wmsc/upload/file/pdf/v2/lcsc/2111120930_Rainsun-microwave-Tech-AN3216-245_C2916136.pdf
But right now I soldered a 31mm wire in its place for testing but I don't get any tx with that either (while rx improved a bit).

[Randy222]

The RF output impedance (impedance matching point) for the ESP32-S3 is 35+j0 Ω, as confirmed by Espressif Systems and detailed in the ESP32-S3 Hardware Design Guidelines.  This value is used as the target for the conjugate matching point in the RF matching circuit to ensure optimal transmission performance, where the S11 parameter should be close to this point.

While the chip's output impedance is 35+j0 Ω, the RF trace impedance connecting to the antenna must be designed for 50 Ω. A π-type (CLC) matching network is essential to bridge the difference between the chip's 35 Ω impedance and the standard 50 Ω antenna system


Custom PCB Designs: The RF pin output impedance is approximately (30+j10)Ω, requiring a pi-type (CLC) matching network (Capacitor-Inductor-Capacitor) to transform the impedance to 50Ω before connecting to the antenna; connecting an antenna directly without this network causes severe mismatch loss.

XIAO_ESP32C6: This board features an RF switch (FM8625H) that allows switching between the onboard and external antenna via GPIO 3 (RF switch power) and GPIO 14 (antenna selection).

Multi-Antenna Systems: The ESP32 supports up to 16 antennas routed through an IO MUX and GPIO matrix, where specific ant_sel_in pins (connected to GPIOs) control external antenna switches to select between up to two active antennas simultaneously.

Modules with Connectors: Boards like the ESP32-S3-WROOM-1U or ESP32-CAM often use IPEX (U.FL) connectors where the internal RF trace is already matched; in some cases (like ESP32-CAM), a 0Ω resistor must be soldered to bypass the onboard antenna and enable the external connector.

[Thalhammer]

Ok so I finally got the new PCBs and assembled them today and while it solved all the other issues (like the missing/wrong reset circuit) I get exactly the same RF issue.
Receiving works fine, but trying to transmit gives absolutely nothing.

I am wondering how I should proceed now. I see two ways
1. Scrap the project in this form and accept the fact that I suck at RF and have no fucking clue what I am doing.
2. Go full in buying a SDR/VNA to debug the issue potentially wasting a bunch of money on the equipment.

I'd really like to figure it out but I feel like doing custom RF PCBs simply is not an option for people without formal training in the field.
Is anyone nearby (Upper palpatine/Bavaria/Germany) with adequate knowledge and equipment that would consider taking a look (obviously with compensation for the time) ?