Sonoff TH10 3.3V instability, ESP8266 hangs after Wi‑Fi connect

User question

instability problems with 3v3 on sonoff th10.
Replaced capacitors but esp hangs and need hard rester. then hangs again few seconds after connecting to network.
further investigations on 3v3 circuit ?

Artificial Intelligence Response

Direct answer to the question

Yes. If a Sonoff TH10 works for a few seconds and then the ESP hangs specifically when Wi‑Fi/network activity starts, the next place to investigate is not only the capacitors, but the entire 3.3 V power path under dynamic load.

Most likely causes, in order:

• 3.3 V regulator/LDO instability or degradation
• Upstream supply sag feeding that regulator
• Missing/cracked local ceramic decoupling near the ESP8266
• 3.3 V droop/noise during Wi‑Fi TX bursts
• External sensor load or board-level solder/trace fault
• Less likely: flash/firmware problem that only appears after network init

The fastest isolation test is:

1. Disconnect mains completely.
2. Power the board from a known-good external 3.3 V supply capable of at least 500 mA continuous and high pulse current.
3. If the unit becomes stable, your fault is almost certainly in the onboard PSU / 3.3 V rail.
4. If it still hangs, suspect ESP8266, flash, boot strapping, or firmware.

---

Detailed problem analysis

1. Why the problem appears at network connect

The ESP8266 is relatively modest at idle, but during Wi‑Fi association and RF transmission it draws short, sharp current bursts. Those bursts are exactly where a marginal 3.3 V rail fails.

Typical symptoms of supply weakness:

• boots normally
• starts Wi‑Fi scan
• connects or almost connects
• hangs, resets, or becomes unresponsive
• hard reset works temporarily
• failure repeats a few seconds later

That pattern is much more consistent with transient supply collapse or regulator oscillation than with “bad capacitors only”.

A multimeter often misses this because it averages the voltage. You need to know whether the rail is:

• dipping below safe operating voltage,
• ringing/oscillating,
• or being polluted by upstream ripple/noise.

---

2. Do not assume the capacitor replacement solved the real fault

Replacing electrolytics is reasonable, but several other faults can produce the same behavior:

A. The regulator itself is weak

If the board uses an AMS1117-class or similar 3.3 V regulator, it may:

• pass DC correctly at light load,
• but fail during pulse load,
• overheat,
• or oscillate with the wrong output capacitor type.

B. The replacement capacitor type may be wrong

This is important:

• Some 1117-family regulators are unhappy with very low ESR ceramic-only output capacitance
• Some general-purpose electrolytics have too much ESR for fast Wi‑Fi current steps
• Long lead lengths also reduce effectiveness

So “new caps” does not automatically mean “stable rail”.

C. The real problem may be upstream

On many Sonoff revisions, the 3.3 V rail is derived from a mains power section, often through an intermediate rail. If that upstream rail sags or carries too much ripple, the 3.3 V stage will fail even if the 3.3 V capacitors are new.

---

3. First diagnostic: external 3.3 V injection

This is the most useful next step.

Method

• Do not do this with mains connected
• Power the logic side only from a clean bench 3.3 V source
• Use a supply that can handle fast current peaks
• A cheap USB-TTL adapter’s 3.3 V pin is often not sufficient

What this test tells you

• Stable on external 3.3 V
  → onboard regulator / upstream supply / decoupling is bad
• Still unstable on external 3.3 V
  → not primarily a 3.3 V generation problem; check flash, ESP, firmware, sensor interface, or boot pins

Important caution

Do not back-feed the onboard regulator carelessly. If possible:

• inject at the ESP/3V3 logic rail with mains removed,
• and verify you are not forcing current backward into another supply section.

---

4. What to measure on the board

If you have a scope, probe these points. If not, you can still do some DC checks, but the scope is far more valuable.

Recommended test points

Test point	What to expect	Fault indication
Regulator input	Stable upstream voltage	Dips or strong ripple during Wi‑Fi
Regulator output / 3.3 V rail	About 3.3 V with small transient dips	Drops below about 3.0 V, oscillation, ringing
Directly at ESP8266 VCC pin	Clean local 3.3 V	Worse droop here than at regulator output means layout/decoupling issue
Ground continuity between regulator and ESP	Very low resistance	Poor solder joint / cracked trace

Scope setup

Probe as close as possible to the ESP8266 supply pins.

Look for:

• voltage droop during Wi‑Fi bursts
• high-frequency oscillation on the regulator output
• 100/120 Hz ripple from the upstream supply
• slow sag that suggests thermal overload

A short ground spring on the probe is much better than a long clip lead.

---

5. What “bad” looks like on 3.3 V

The ESP8266 does not tolerate supply collapse gracefully. Undervoltage can cause:

• spontaneous resets,
• flash read/write errors,
• watchdog resets,
• lockups that look like firmware hangs.

Red flags:

• 3.3 V dropping by more than roughly 150–250 mV during network connect
• rail going below roughly 3.0 V
• oscillation or ringing superimposed on the rail
• regulator getting too hot to touch quickly

Even if the average voltage reads 3.28 V on a DMM, very short dips can still crash the ESP.

---

6. Check the regulator, not just the capacitors

Depending on TH10 revision, identify the actual regulator fitted. Do not assume the exact part number from another board revision.

Inspect:

• package marking
• solder quality on all regulator pins
• input and output capacitor placement
• local ceramics near the regulator and ESP

Regulator failure modes

• internal pass element weakened
• thermal shutdown or thermal foldback
• unstable loop due to wrong output capacitor behavior
• excessive dropout because upstream voltage is collapsing

Practical test

• Measure regulator input and output simultaneously during Wi‑Fi connect
• If input is stable but output collapses: regulator or output decoupling problem
• If both collapse: upstream PSU problem

---

7. Very important: inspect small ceramic capacitors near the ESP8266

This is often missed.

Electrolytics handle bulk energy, but the small MLCCs near the ESP handle high-frequency current pulses.

If one of these is:

• cracked,
• missing,
• poorly soldered,
• or detached by rework heat,

the board can appear “mostly fine” but fail during RF events.

Good practical experiment

Temporarily add, physically close to the ESP8266 supply pins:

• 100 nF ceramic
• 4.7 µF to 10 µF ceramic
• optionally 47–100 µF low-ESR bulk capacitor

If stability improves, the issue is local decoupling / impedance of the 3.3 V distribution path.

---

8. Upstream supply investigation

If the TH10 is stable on external 3.3 V but not on its onboard supply, the next stage to inspect is the rail feeding the 3.3 V regulator.

Depending on board revision, this may be:

• a small off-line SMPS section,
• or another derived DC rail feeding the logic regulator.

Check for:

• poor bulk capacitance on the upstream rail
• high ripple
• tired primary/secondary capacitors
• bad optocoupler / feedback behavior in the PSU
• cracked solder around transformer, diode, regulator, or module pins

Field pattern seen on older Sonoff boards

Some older Sonoff hardware revisions have been reported to behave properly when powered from serial/bench 3.3 V, but become unstable on mains-derived power. That strongly points to:

• poor onboard decoupling,
• noisy supply,
• or marginal regulator behavior under RF load.

That does not prove every TH10 has the same fault, but it is consistent with your symptoms.

---

9. Disconnect everything non-essential

Before deeper conclusions, test the board in its bare minimum configuration.

Disconnect:

• external temperature/humidity sensor
• anything on GPIOs
• any external serial adapter that might be loading pins
• relay load, if practical for safe testing

Why this matters:

• sensor cable or connector contamination can pull the rail down
• some peripherals can disturb boot strap pins
• extra load on 3.3 V may only become critical during Wi‑Fi bursts

If the problem disappears with the sensor unplugged, inspect:

• the sensor itself,
• cable damage,
• jack corrosion,
• and leakage around the input protection network.

---

10. Do not ignore solder joints and copper integrity

Intermittent or resistive joints on a low-voltage high-pulse rail can create exactly this problem.

Inspect under magnification:

• regulator pins
• capacitor pads
• ESP module castellations
• upstream PSU pins
• connector joints
• any reworked area from the capacitor replacement

A hairline crack in the 3.3 V path can look fine with a meter at idle but fail under burst current.

---

11. Separate hardware power issues from firmware/flash issues

If the rail looks clean and the board still hangs after joining the network, consider:

• flash corruption
• bad flash soldering
• firmware bug after Wi‑Fi init
• watchdog reset due to software task starvation
• boot mode / GPIO strap issue

Very useful test

Capture serial output during boot and failure.

On ESP8266:

• boot log at 74880 baud
• firmware logs usually at 115200 or configured value

If you see repeated resets or watchdog traces, note them.

Minimal firmware test

Flash a very simple firmware that:

• boots,
• connects to Wi‑Fi,
• does nothing else,
• does not drive sensors or MQTT.

If minimal firmware is stable on clean external 3.3 V, but your normal firmware is not, the problem may be software load or flash access timing rather than pure hardware.

---

Current information and trends

Based on the online sample answers and current field experience with older Sonoff-class ESP8266 products:

• A recurring real-world symptom is:
  works on bench/serial 3.3 V, fails on mains power
• Older boards can be more sensitive to:
  • marginal decoupling,
  • regulator choice,
  • and firmware that increases Wi‑Fi activity or CPU load
• Modern embedded designs generally avoid weak 1117-style regulator implementations for RF MCUs unless:
  • layout is good,
  • local ceramics are close,
  • and the upstream rail is clean
• Current best practice for Wi‑Fi MCU boards is:
  • low-impedance local supply network,
  • short return paths,
  • better transient response,
  • and often a switching regulator or a well-characterized LDO rather than a generic clone

One important correction to some common advice:
it is risky to treat every “ESP hangs on Wi‑Fi” case as purely firmware. In many Sonoff-style products, supply integrity remains the first thing to prove or eliminate.

---

Supporting explanations and details

Why Wi‑Fi causes trouble while simple boot does not

Think of the ESP8266 like a load that is calm most of the time, then suddenly asks for a burst of current in a very short time. A supply can look fine on average but still fail in those moments.

Equivalent engineering view:

• the regulator has finite loop bandwidth,
• the capacitors provide transient current until the regulator catches up,
• ESR/ESL and trace impedance matter,
• if the combined impedance is too high, VCC dips,
• the ESP misbehaves.

Why the replacement capacitor type matters

A regulator is part of a control loop. The output capacitor is not just a reservoir; it is part of the loop compensation environment.

So the wrong combination of:

• capacitance,
• ESR,
• ESR versus frequency,
• and physical placement

can make the regulator oscillate or respond poorly.

Why a DMM is often misleading

A DMM might show 3.30 V while the actual rail is doing this:

• 3.30 V average
• but with short drops to 2.85 V during TX bursts

Those short drops are enough to crash the ESP.

---

Ethical and legal aspects

Safety

This is the most important non-technical point.

• The TH10 is a mains-powered device
• Primary-side circuitry can carry lethal voltage
• Even if the low-voltage section is isolated on your revision, the board still contains hazardous areas

Best practice:

• debug logic power with mains removed
• use a bench supply for low-voltage tests
• avoid probing live mains sections unless you are equipped and trained
• do not defeat isolation or enclosure safety barriers permanently

Regulatory / compliance considerations

If you modify the power architecture substantially, the device may no longer match its original compliance status. For permanent installation:

• preserve creepage/clearance
• use properly rated components
• secure any added wiring/capacitors mechanically
• avoid exposed low-voltage modifications inside a mains enclosure

---

Practical guidelines

Recommended troubleshooting order

1. Test with external clean 3.3 V

   • If stable, onboard supply is the problem.

2. Scope the 3.3 V rail at the ESP

   • Check during Wi‑Fi connect.

3. Scope the regulator input

   • If input collapses too, fault is upstream.

4. Inspect/replace the 3.3 V regulator

   • Especially if output is unstable with good input.

5. Check small ceramics near ESP and regulator

   • Replace cracked/missing MLCCs.

6. Add temporary local decoupling at the ESP

   • 100 nF + 10 µF close to the module.

7. Disconnect the external sensor

   • Rule out extra load or leakage.

8. Check serial logs / reset behavior

   • Separate power fault from firmware/flash issue.

9. Reflash minimal firmware

   • Confirm whether the hang is load-related in software.

---

Best practices for implementation

• Use short wiring when injecting external 3.3 V
• Use a supply with good transient response
• Probe directly at the ESP, not only at the regulator
• Rework with proper flux and inspect under magnification
• If replacing the regulator, use a known genuine part or a well-behaved equivalent
• Do not rely on FTDI 3.3 V power for final proof unless you know it can handle the peaks

---

Potential challenges and how to overcome them

Challenge: “Voltage reads fine on my meter”

• Use a scope
• Or at minimum use min/max capture, though it may still miss fast dips

Challenge: “I replaced all electrolytics already”

• Check ceramics
• Check regulator
• Check upstream ripple
• Check solder and trace integrity

Challenge: “It only fails when on mains”

• Strong evidence for onboard PSU noise/sag
• Compare with external 3.3 V supply test

Challenge: “It hangs, not resets”

• Undervoltage on ESP8266 can produce lockups, not only clean resets
• Do not dismiss power just because it is not rebooting cleanly

---

Possible disclaimers or additional notes

• TH10 internal power topology can vary by hardware revision.
  Do not assume the exact schematic from another Sonoff board.
• Some generic advice online overstates exact thresholds and current values.
  Use them as guidance, not as guaranteed numbers for every board.
• Some 1117-family compatibility advice is clone-dependent.
  The safe engineering approach is: check the actual regulator and its recommended capacitor behavior.
• If the board is old and has been heat-stressed, both the regulator and the ESP flash may be degraded at the same time.

---

Suggestions for further research

If you want to go deeper, the most useful next data would be:

• exact TH10 hardware revision marking
• clear PCB photos, front and back
• regulator part marking
• scope capture of:
  • 3.3 V at ESP VCC
  • regulator input
  • timing around Wi‑Fi connect
• serial reset log
• whether it is stable on external 3.3 V
• whether the sensor is connected during the test

If you provide those, the fault can usually be narrowed down very quickly.

---

Brief summary

Your symptom strongly suggests a 3.3 V rail integrity problem under Wi‑Fi burst load, not just “bad capacitors”.

The highest-value next steps are:

• power it from a clean external 3.3 V supply with mains disconnected
• scope the 3.3 V rail at the ESP during Wi‑Fi connect
• check the regulator and the upstream supply
• inspect/replace local ceramic decoupling
• test with the sensor unplugged

If external 3.3 V fixes it, the board PSU/regulator is the fault.
If not, move to ESP/flash/firmware investigation.

If you want, I can give you a pin-by-pin measurement checklist for the TH10, or help you interpret scope traces / serial reset messages.

Disclaimer: The responses provided by artificial intelligence (language model) may be inaccurate and misleading. Elektroda is not responsible for the accuracy, reliability, or completeness of the presented information. All responses should be verified by the user.