https://community.element14.com/technologies/internet-of-things/The Internet of Things (#IoT) becomes a popular industry topic and data illustrating perspectives are very impressive: IHS predicts 29.2 billion of connected devices by 2020. Huge opportunity! Let's discuss about technology, market trends and solutionsen-USTelligent Community 12https://community.element14.com/technologies/internet-of-things/f/forum/57202/on-device-fine-tuning-in-608-bytes-gravoptmini-optimizer-w-twin-health-monitor-for-stm32Mon, 17 Aug 2026 23:27:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:b876e23a-f8bb-4ad7-ae88-42497b6bde00regdredBuilt a minimal on-device training stack for STM32 — fits in 608 bytes RAM total. The problem: Adam optimizer needs exp_avg + exp_avg_sq — 2× parameter RAM. On STM32F103 (20KB SRAM) with a 5K parameter model that's 40KB just for optimizer state. Doesn't fit. What I built: GravOptMini — momentum optimizer with quantile-based gradient freeze. Only exp_avg (1× param RAM). Freezes low-gradient parameters to reduce compute. W-Twin Lite — power-law baseline fitted on early steps, tracks deviation. Pure float32, no scipy, no numpy. Rolling window of 20 values. ~240 bytes. Benchmark (Python simulation, STM32 memory constraints): - Model: 46 params, 184 bytes - GravOptMini state: 368 bytes (exp_avg only) - W-Twin Lite: 240 bytes - Total: 608 bytes - MNIST accuracy: 93.2% vs Adam 97.4% - W-Twin overhead: 3.9% per step - Detection: anomaly caught 96 steps before visible loss divergence, 0 false alarms on clean run Next step is a real STM32 test with physical hardware. C port of both components is the plan. Has anyone done on-device fine-tuning on STM32? Curious what optimizer you used and what constraints you hit. Code: github.com/Kretski/WTwinedge aitinymlembeddedsystemsstm32optimizermachine learningioton-device learninghttps://community.element14.com/technologies/internet-of-things/b/blog/posts/mspm0-opamp-interconnectionFri, 14 Aug 2026 11:04:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:2543e862-7cae-4a19-8faa-b7d77ad69d6dJan CumpsSeveral MSPM0 controllers have more than one OpAmp on board. There are MUX settings to interconnect them. I couldn't find a generic picture of how two OpAmps can be wired up, so I made one. It's based on Figure 17-8. Two OPAs in Difference Amplifier Configuration of Texas Instruments Technical Reference Manual for the MSPM0 L-Series 32-MHz Microcontrollers. click for full size Some connections aren't drawn, but have a label name. Comment if you 'd like to see some of them as a line.MSPM0https://community.element14.com/technologies/internet-of-things/b/blog/posts/mspm0-sample-photo-diode-with-internal-opamp-and-adcFri, 14 Aug 2026 08:30:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:1d6f8951-2328-4895-aaf8-ae634902ac9bJan CumpsWhen you try to measure light intensity with a photodiode, you have to build some circuitry around it. There are several ways to do that. In this post, I'm investigating TI's application note MSPM0 Subsystem Design Transimpedance Amplifier . This is an interesting circuit. Because the OpAmp clamps the diode voltage to 0 V (same value as positive input), the only way the energy in the diode can escape, is as current. That current has to come from the output, via the feedback resistor (it can't come from the OpAmp input, because that's the same level as the other side of the diode: ground). That also means that the output has to be a higher voltage than ground: ground + (current * resistance). And that's something the ADC can then measure. See here for a better explanation. The OpAmp used in this design is one of the two that are embedded in the MSPM0 silicon. They are rail-to-rail chopper types (TI calls them zero-drift, zero-crossover chopper operational amplifiers). The microcontroller gives you control for: multiplexing in- and outputs (to pins, references, DAC, ADC, each other) a set of voltage reference sources to use (including that DAC) optionally a number of preset gains bandwidth, chopper on, rail-to-rail on, ... In the photodiode design, we use the OpAmp in a very classic way: the in- and outputs are routed to external pins chopper off programmable gain off rail-to-rail off high bandwidth (TI uses different setting in the transimpedance example and the out-of-box source. I used the settings of the latter) The feedback loop is built up by external components (see schematic drawing at the top of this post). TI has done the homework to specify them based on the specs of the Vishay VEMD4110X01 that's mounted on the launchpad. The capacitance of the diode is dependent on the voltage over it. In our case, there is 0V over the diode, and capacitance in that case is 5 pF. The max current is in the low single digit µA. Let's assume that the diode generates 1 µA. With the 2.37 MΩ resistor in the feedback loop, that would give an OpAmp output of 2.37 V. The rest of the design is uneventful. The MSPM0 ADC input is muxed to the same pin that's connected to the OpAmp output (there is an option to mux it internally on silicon too). The reference voltage is VDDA (analog VDD, on this launchpad the common 3.3 V). The ADC is set to single shot mode, and gets triggered by one of the timers. That's all.MSPM0photonhttps://community.element14.com/technologies/internet-of-things/b/blog/posts/august-12th-eclipse-can-i-measure-itTue, 11 Aug 2026 09:27:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:68745365-b70e-4970-bdc4-5441f5cedf2dJan CumpsTomorrow there's a solar eclipse. In Brussels, it's a partial one. 90%. I'm going to try and measure the light intensity during that event. image source: https://www.timeanddate.com/eclipse/in/belgium/brussels?iso=20260812 Google AI: In Brussels, the Moon will obscure 89.5% of the Sun’s surface (with a peak magnitude of 0.911 ) during tomorrow's solar eclipse on Wednesday, August 12, 2026 . Exact Local Timeline for Brussels (CEST) Because the eclipse occurs right before sunset, timing is critical: 7:18 PM – Eclipse Begins : The Moon first touches the edge of the Sun. 8:13 PM – Maximum Coverage : The peak of the deep partial eclipse occurs. 9:09 PM – Eclipse Ends : The event concludes just minutes after the Sun dips below the horizon. I'm going to use TI MSPM0 Launchpad. The kit has a light sensitive diode. And the controller (MSPM0L1306) has a good analogue front end. Current status: do everything :). I don't have the code yet. But I did check out the schematics for the light sensor. What I'll probably use: that light sensitive diode one of the OpAmps available on the L1306 ADC Timer UART I'll have to use the OpAmp as Transimpedance Amplifier . It 'll also have to bias the LED circuit. Luckily, there are examples . I have done this exercise before in 2015, with a Hercules Launchpad. That time, I also took a photo of the sun every so many minutes. It was a clouded day... If I manage to get my design working this evening, I'll share the findings in our community. Update the day after: it worked. Berfore the eclipse stated, I put my launchpad in a location that 's not directly in the sun: I made a design based on TI's Out Of Box example and that application note for transimpedance amplifiers. It did an excellent job - although the photodiode has a daylight blocking filter. There is a measurable component of IR in sunlight, so we are able to measure that. Here is a graph showing the measurements. start: 19:05 end 21:15 The eclipse happened when the sun was about to set. That's why, after the eclipse, you see both the effect of the end of the eclipse, and of the sun dropping behind the horizon. Google AI: In indirect sunlight, the percentage of energy between 800 nm and 1000 nm ranges from 4% to 13% of the total spectral energy . Widening the wavelength window to 200 nm captures a massive portion of the near-infrared (NIR) spectrum. In standard, direct global sunlight, this specific band is highly energetic, accounting for exactly 15% of the sun's total energy (providing about 150 W/m² out of the standard 1,000 W/m² solar baseline). However, just like the narrower band, indirect atmospheric scattering changes these numbers. 1. Clear Sky Indirect Sunlight (Open Shade): ~4% to 6% If you are standing in the shade on a perfectly crisp, cloudless day, your light is coming entirely from the blue sky. The Atmospheric Filter: Because Rayleigh scattering only affects shorter wavelengths (UV and blue light), the long 800–1000 nm infrared waves pass straight through the atmosphere without bouncing into the shade. The Result: The light is heavily depleted of infrared. Even though the window is wide, the 800–1000 nm band only makes up a small fraction ( ~4% to 6% ) of the ambient shade light. In the comments below, I posted a few action photos. Thank you for reading along.MSPM0eclipsesolarlaunchpadhttps://community.element14.com/technologies/internet-of-things/f/forum/57151/i-built-micropidash-real-time-web-dashboard-in-under-20-lines-of-micropython-no-cloud-no-frameworkTue, 04 Aug 2026 17:52:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:94854de6-6644-406c-aae4-1db42b05f13eKritishBeen building IoT projects every day for my #100DaysOfIoT challenge and kept running into the same problem — monitoring sensor data from ESP32/Pico 2W in a browser was always a mess. So I built micropidash. real-time web dashboard in under 20 lines of MicroPython. No cloud, no framework. Just shipped v2.0.0 with live sensor graphs — tested with DHT11 on Pico 2W, temp + humidity updating in the browser over WiFi. pip install micropidash github.com/kritishmohapatra/micropidash Would love feedback if you try it!element14 communitymicropythonraspberry pihttps://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237530Fri, 24 Jul 2026 12:29:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:d0a7e6f8-86c1-43ec-91de-42f16d3e983bmichaelkellettThe FP6277 is a boost converter - not a battery charger. You don't need a booster to charge a Lithium Ion battery from USB - just a charger control chip. https://www.adafruit.com/product/259 https://www.amazon.co.uk/Lithium-Battery-Charging-Protection-Functions/dp/B08DQXPNR1/ref=asc_df_B08DQXPNR1 You could then connect the Pico directly to the battery. https://randomnerdtutorials.com/power-raspberry-pi-pico-6-different-ways/#batteries I haven't tried or tested any of the above. I suggest that you do a good bit more web research before spending money. MKhttps://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237529Fri, 24 Jul 2026 11:27:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:d569f81e-89aa-444c-b316-86fe353b57c4GeoForemanHi, Sorry for being a noob. I want to start with the charging unit, I am using FP6277 as charger / booster, I want to connect the 3.7v 4000mA battery to the FP6277 and raspberry pi pico 0 : RP2040, in such a way that it it can charge the battery while connected to usb port and run and boost the voltage as necessary even without connected to USB. It would be helpful if you can send me a schematics so that I would have an idea. I need to see the connection how it is connected also to pico ports. Thanks.https://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237528Fri, 24 Jul 2026 11:19:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:a0ebe7c2-23c4-48ac-bb83-0a77643ff39dGeoForemanThanks I will draw it.https://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237527Fri, 24 Jul 2026 09:59:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:52adb745-d369-40f2-bffe-dc964990c098dang74Thanks dougwhttps://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237526Fri, 24 Jul 2026 09:58:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:0e53e317-cff4-4e11-af1c-5fff82a8e62cdang74Thanks michaelkellett this is what I want to achieve.https://community.element14.com/technologies/internet-of-things/f/forum/57122/re-power-connection-problem/237523Fri, 24 Jul 2026 08:45:00 GMT93d5dcb4-84c2-446f-b2cb-99731719e767:b88d0e91-d968-4fd1-baf8-3f036a89d04emichaelkellettIf you must have 3.3V you will have a hard choice. To get the most out of a nominal 3.7V Lithium Ion battery you would like to run it down to 3V or even 2.7V. Fully charged it will deliver 4V. For low power stuff you could use an LTC3532 regulator which will do the business: I think this is a quite common approach - and there will be other (and cheaper) chips than the LTC3532 but I've used it and know it works. It is also supported by a good model in LTSpice. The max input voltage for the chip is only 5.5V so don't forget to have some kind of transient clamp on the input. MK