yup. I felt a little spoofed. With that inaccuracy it seems that an external ADC isn't really that much better than a software-compensated ESP32 ADC. However, the precision of an external ADC could be interesting for scaled up projects: you are trading the extra costs for an ADC to skip the individual calibration process of each ESP32 ADC.
@@tigerresearch2665 The external ADC only performs so bad due to his own "correction formula" that very likely only corrects errors on the analog end (aka from V_ref to the PGA inside the ADC). So it shows the correct value at this one point, but not anywhere else. One error that affects the analog end, which I would suspect here: High sampling rate -> low input impedance -> voltage drop Correcting ADC values in code like he does is not something anyone should do. The first step is to correct the hardware/setup! Let me compare that to a car that constantly steers to the left: Do you just "counter steer" to make it straight... or do you fix your car so it drives straight?
I am using an ESP32 to monitor the battery voltage on my boat and send an email if it gets too low. I found the measurements very noisy due to occasional glitches. My solution was to take 8 readings and ignore the largest and smallest and average the middle 6. I do this twice and average them. It is quite easy to do in the code as you just have to keep track of the largest and smallest measurements and subtract them from the sum and divide by 6. I found this better than longer averages so that I did not get multiple glitches in my sample period. The result is very stable and quite fast. I very much enjoy your videos I am a retired circuit designer. I designed test equipment for HP (Now Keysight) and things like cable modems in the early days. Now it is a hobby.
Good idea because you can eliminate the far glitches. I did not investigate the distribution of the measured values, but maybe they have a pattern you break with your formula. I like my Keysight MSOX3045 ;-) Good products!
@@AndreasSpiess The distribution was not Gaussian. It was mostly correct values with a few outliers. It is actually a pretty standard way of filtering. There is even a name for it but I can't find it any longer. I learned it from a statistician who worked for me for a short time.
Thanks for this video! I've been puzzling over how to reduce the noise of a voltage reading for a project without losing responsiveness introduced by filtering. Based on your video, I switched over to an external ADC (ADS1115) and it solved the problem. Much appreciated!
Sounds neat, were they hand-matched R2R ladders? I always wonder why you can’t just buy R2R resistor networks in single packages, like a SIP10. Though these days I’d be tempted to just use a half-bridge and low-pass filter, like a good-old class-d amp.
The AVR ATmega328p used in the Arduino, and other AVR chips, have a special sleep mode for ADC noise reduction. It keeps the ADC running, but stops all the digital processing. See the "ADC Noise Reduction Mode" in the datasheet. It could be that the external ADC chips appear better simply because they are physically further away from the running digital logic.
Interesting. I didn't know Arduino had a sleep mode.
4 роки тому+4
@@Slutuppnu I don't know about the Arduino library, but with AVR-GCC you have fine-grained control over it. www.nongnu.org/avr-libc/user-manual/group__avr__sleep.html
4 роки тому+8
PS: The Arduino IDE already uses AVR-GCC, so you can use the functions that I linked from Arduino projects as well.
@@Slutuppnu Most AVR chips (the ones inside arduinos) don't have just one sleep mode but a few of them. It has (as has been said), an ADC noise reduction in which the CPU and some other digital parts of the chip are turned off in order to remove noise from the ADC signal; it has a CPU sleep mode that allows you to save a bit of power during delays or when waiting for a short timer to wake you up; and finally it has a deep sleep mode, in which the CPU, fast clock signal (either the internal fast oscillator or the external crystal oscillator), and most peripherals are turned off, except for the WatchDog Timer (low power timer) and a few more things like pin interrupt wakeups. This allows for very low power applications, in which you can put the chip to sleep and only use a few microamps of current, which for battery applications is really useful. For those types of applications, you have to take care and desolder all the LEDs on the board that you can't control, make sure to avoid using any LDOs onboard, etc.
The ADS1115 is my favorite ADC for microcontrollers. I started using it when I moved over to 32bit microcontrollers, but needed a way to read 5V automotive sensors (using a Teensy 3.2). I had 4 sensors (an MPX5700AP pressure sensor, an NTC thermistor, an AEM Wideband oxygen sensor and a 0-100psi oil pressure sensor), the ADS1115 had 4 inputs, it was a match! I didn't have to average values to get a stable reading, like you said, the stability of the ADC is fantastic.
Awesome. Thank you for the great content! Just my two cents: 1. I've had to learn the hard way, that many (all?) of the ADC that have a MUX only have a single S/H aufter the MUX. This leads to problems, when reading signals on different channels with very different voltages in quick succession. There is quite a bit of "crosstalk". It's really important to make sure that the inputs to such an ADC are low impedance. 2. When I do averaging on a uC, I do 2^n samles (8, 16, 32...1024) samples, add the read values up (make sure to prevent overflow) and just use shift for the division at the end. Especially on uC without an FPU this speeds the averaging up quite a bit.
since the input is in parallel to the low-side resistor of the voltage divider only its value really matters - well actually it is of course the parallel-combination of the upper and lower resistance in the voltage divider... And for the ATmega328 the input resistance of the ADC-channels is astonishingly high! I tested it with discharging a 1µF capacitor from 5 V to zero through a 100 MΩ resistor - it took about 15 minutes to discharge and the resulting curve indicates an input resistance beyond 1 GΩ.
Uwe Zimmermann I was going to ask what you used to measure that voltage, but I guess you were just using the ADC itself. I imagine most of the leakage current will be through the protection diodes, if it had none then it would be a real recipe for ESD. Which reminds me of a friend using his atmega ADC for measuring back-emf of a solenoid. Directly on the 12V half-bridge output that was powering the solenoid. With no extra diode or resistors. The arduino got really hot, but somehow the circuit actually worked.
@@uwezimmermann5427 I am not talking Any specific microprocesor. I had an adc in a low power circuit where I wanted to messure the battery voltage, here I used a resistor divider in 10 Mohm Range to avoid draning the battery. In this applikation the messured voltage was wrong due to current flowing into the mcu.
If you tweak the adc settings in the ESP32 down to 10 bits, and some other changes, the noise and accuracy gets a LOT better, I’m not at my computer at the moment but I will post the code when I get a chance.
Here is the code of the @@TheDefpom comment: "Here you go, here is the code to help get the ESP32 ADC to behave better: // configure ADC for getting battery voltage pinMode(VoltageSense, INPUT); // Battery Voltage Sense input analogSetPinAttenuation(VoltageSense, ADC_0db); // set the attenuator for the ADC to allow 1.1V input range analogSetWidth(11); // set the resolution width to be 2048 instead of the default 4096 to try and reduce noise analogReadResolution(11); // set the resolution to be 2048 instead of the default 4096 to try and reduce noise"
Greetings from South Africa Andreas. Well done as usual! I am always surprised how much information you share with great clarity and structure in such a short video clip. When I learnt this stuff (long before internet), choices were fewer and one had to study the data sheets thoroughly (or copy somebody's example from Elektor, EE, ETI etc). Microcontrollers (with built in ADCs) only arrived later and choices were fewer - which was a blessing, because obsolescence took longer. Hobbyists today should be thankful that you share years of experience, crammed into a nutshell...and for free. Love your work and your very structured method of teaching (unlike my ADHD brain which likes to skip around). Best Regards, Allen.
Very interesting video as always, thank you. Another benefit of using the ADS1115 is that the internal PGA can be configured to read a differential signal, this is very useful for measuring some sensors, although it also results in a slower data acquisition rate.
This trick can also be used in general by the Arduino by using two analog inputs connected to the signal direct and the signal amplified by a fixed op amp.
Hints for those starting out -- for built-in ADCs, throw away the bottom bit as it's most likely useless anyway. You can average samples equal to a power of 2 by adding the integer values and doing a right shift -- for example add up 8 integer ADC samples and right shift 3, a useful trick when your microprocessor doesn't have a divide instruction! What's your signal bandwidth? Adding a single pole RC low pass filter in front of the ADC can help. And do not expect the input impedance of the ADC to be constant.
Another great video from a very competent presenter. Just a point for a possible misconception: there are two other ADC architectures in common use, namely pipelined and sigma-delta. The first tries to keep the number of parallel comparators required in check by converting the output of a comparator stage back to analogue using a DAC, subtracting this output from the original input, and feeding the result to another comparators' stage. Three stages are quite common with typically 4bits per stage; leaving aside the unavoidable latency the conversion takes one clock cycle like the comparators' ADC. This breaks for high resolutions because analog subtractions accumulate too much error. Then sigma-delta comes to the rescue where in fact a control system tries to get its 1 bit output bitstream average as close to the input as possible (its operating principle is not easy to understand in detail though); 24 bits and higher resolutions became possible. Successive approximation is favoured by the MCU manufacturers for a very simple reason - it requires a single analog comparator and an R-2R DAC - cheap, accurate and reproducible architecture that covers nearly all practical DAC needs. Moreover it can be difficult to find a standalone successive approximation ADC as most are either pipelined (higher sampling rates) or sigma-delta (for high, eg audio, resolution). Again for a typical maker these are too finer points but I wanted to make sure they are mentioned in the footnote. Thanks again Andreas for your work
Thanks Andreas for this interesting video. I've used ADC lots of times, and never had concerns about the quality of the conversion (noise reduction with average method). In a current project, getting feedback from the servo's potentiometers for a robotic arm, I've used a calibration table, but now I'm going to introduce your formulas. Thanks again, Juan.
It depends on the desired accuracy. I compared, of course, with the data sheet.Calibration tables are good if the ADC is dependable wrong and not stochastically wrong...
Watched already a lot of videos of your channel and at the beginning it was a lot of „häääää, versteh grad überhaupt nix“ but from one video to the next I start to understand what you are explaining us. Have to say that I am a totally new beginner. My goal is to measure the 2 12V Solar Batteries in the OFF-Grid garden of my parents in law. Tried a lot with the ADC from the ESP32 itself and as u mentioned it’s readings are just bullshit. Thank you for introducing the ADS1115 to me. I guess that will fix my measuring problem and can finally bring the project into production. As I am original from Vorarlberg and worked many years in Kanton SG I love to hear your swiss accent as it remembers me of that great time in my life! Electronic wise you are my personal hero ;-)
For this purpose you probably could also use a INA226 as suggested by other commenters. And I am glad you learn something with each video. We all started as beginners in this field.
This is such an important topic. I send a million thanks and encourage this topic. Getting signals INTO the computer is probably just as important as having the computer to begin with. This topic makes me consider re-writing all of computer science just to make an analog computer.
Hi Andreas! Great video as always! One note about the ADS1115 is that it is actually a 15-bit signed differential ADC( AC !!! ), or 16-bit single-ended. Differential measurements offer more immunity from electromagnetic noise. This is useful when using long signal wires or operating in an electrically noisy environment. I have been using an ADS1115 to measure the amperage of our utility inlet with a sct-013. The fine art of accuracy I've seen is matching the sampling rate of the adc to the frequency of your local provider to sample the highs and lows of the sine wave as accurately as possible (50Hz here in ZA).
You are right with differential measurements. Concerning 50Hz: You have to synchronize because the ADS1115 is slow. A diode and a RC filter probably would help to get DC out of AC
@@AndreasSpiess on the sct013 each 10A represents 1v. Thus getting Amperage readings are easy and safe with non invasive clamps. Only when it comes to the measurement of voltage, territory gets unsafe. I have opted for a 220v primary, 12v secondary to convert ac to DC proportionally followed with a voltage divider and capacitor. With both voltage and amperage the apparent usage can be calculated to identify power consumption patterns at home. I cannot thank you enough for the extensive and concise research you have given to this community! You inspire us!
You have confirmed my discovery. Thanks! Unfortunate that we have to rely on "small and casual" posts or trial&error methods to correctly use the esp32.
Thank you for the breadth of the experiments to help reinforce the lessons. Seeing how externally sourced ~5v can impact the readings was an "of course!" moment, since I'm used to using a good regulator versus an adjustable power supply for projects. Spotting things like how the radios on the chip may affect the ADC are very good clues. I'll investigate some of the external ADCs, since I'm thinking about a project that will use a mesh radio network and I had planned to use analog inputs as well... thanks again!
I'm amazed every time I see your great videos!!! You are testing stuff that went true testing before manufacture, but you made it better - why isn't Arduino listening??? If Arduino and ESP's had better ADC, the cost would not be that much higher - right? But now we can use your recipe and suggestions! Thank you again Andreas. And wishing your finger to heal FASTER!
That was very enlightening, too bad my favorite MC ESP32 yielded such weak (lab) results - still plenty accurate though for many projects. Will look into the ADS1115. Dankeschön !
Great video. One thing to note. If you use an external i2c ADC, you can not set up any sort of interrupt driving timing to collect sames in the background at a given rate since i2c does not work within a ISR. This leads to a bit of work to make sure loop() gets called at a reasonable rate.... Just something to be aware of.
Neat video! I’d have liked it more if you went over the details of successive approximation registers, and how a delta-sigma ADC (or DAC) works, but maybe that’s outside the scope of this video. On the topic of ADCs and DACs, I’d be interested in a video on DSP algorithms being implemented in an arduino, be it for audio or for filtering out narrow-band noise. Also when you said that SDRs use these very high speed GHz ADCs, I don’t think they actually do. They use lower speed ADCs with mixers and tuneable local oscillators to generate the I and Q signals. Direct sampling SDRs are very rare. I’d also have mentioned the fact that the ADCs have a finite input impedance, so putting a network of resistors in front of one without buffering it after could cause some systematic error. Well, you often use an instrumentation amplifier there anyhow so maybe it’s not too useful of a tip.
1. This channels is for makers, so the most important decision is what not to show. 2. I define SDR not only as "SDR receivers for the general public". I define them as "digitalize the signal and do the processing in software". So radar and many other applications fit my definition. 3. You are right with the input resistance. If you calibrate the readings I assume you can live witout an instrumentation amplifier for 10 or 12 bit resolution. For more precise applications you are right, of course
The formula @Andreas Spiess is using is not the usual average formula. He uses a so called filter with feed back and I don't know why he is calling it average but that one for sure is not just an average. And the second point when he says "when I use 10 samples or even 100 samples in that formula" .. those are not the actual samples. That number tells how much we should fed back to the initial value so basically to explain it we have an initial x=0.551 and in the loop he is calculating the next x by taking 99 parts of original x value (0.551) and adding 1 part of the new analogRead value thus it basically means we are just drifting to the left or right of the initial value and the amount of drift is expressed in the ratio we are using 1/10, 1/100, 1/1000 and so on parts of the new value mixed back to the initial one. Also this number will be the speed of conversion to the initial value so the higher value we use the slower is the speed and the more analog reads we need to go back to normal and at the same time the less is the chance to affect the final number by the new erroneous reading. Maybe @Andreas Spiess does not fully understand the formula he is using which I doubt but who knows maybe he can explain it in some further videos.
@@valcob6617 Also the filter is prone to accumulating round off error for high filtering coefficients and my guess is this is why he gets 10% error on ESP32's ADC
I had to chuckle a little about the example of measuring a 24V Solar panel and then setting the range to 30V (around 13:23 in your video). A solar panel that you would use with 24V batteries in a real-world scenario would normally be a 72-Cell panel, which typically has an open circuit voltage around 45V @25ºC, and quite a bit higher when it gets really cold. I realise it was just an example in your video, but for anyone wanting to really measure a "24V Solar Panel", maybe increase that range in your resistor divider a little :-) As usual, thanks for an interesting Video Andreas - toll!
Thanks for another great video! In researching my gas sensor project, I came across a method people say improves the accuracy of the readings, though I don't really understand it. You apply this calculation to the reading on the ADC pin: volts = -0.000000000000016 * pow(reading,4) + 0.000000000118171 * pow(reading,3)- 0.000000301211691 * pow(reading,2)+ 0.001109019271794 * reading + 0.034143524634089; (Also note that there is a more efficient way to do this calculation (10x faster), but it yields a slightly different result, so I'm leaving the original one here.)
For some reason the calculation got scrambled when I pasted it in to the comment - apparently having a "+" after a "-" gets interpreted as strike thru... Hopefully you can figure it out.
@@AndreasSpiess I forget how I originally found the calculation, but if you Google "ESP32 ADC Polynomial" you will find a lot of people use it to make the values returned by the ADC more accurate. You noted that the ESP32 wasn't accurate, but apparently it can be improved with this technique. I'm happy with the results I'm getting, but my application doesn't need it to be extremely accurate.
Nice explanation on the ADC basics. This can be confusing for beginners and experienced people alike. The resistor values used when measuring higher voltages(30V) are way to high. With a few meg ohms the ADC's sampling capacitor cant fully charge and your results will be off. As the voltage drops the problem will get worse. You need to either increase the sampling time (not rate) or put a big cap (10-100nf should do) at the junction of the voltage divider. The capacitor will then supply the required charge to charge up the ADC's sampling capacitor. This will obviously mean the sampling RATE will be dependent on the RC values of the voltage divider and the cap. If you do the same experiment with resistor values in the range of 10 - 100K as the voltage divider , you will get much better results. This is not an issue with the ESP32 , the same will hold true for all ADC's. I use the STM32 a lot and have seen this exact issue. If you have to sample a high impedance source you will have to increase your sampling time. You will also see more noise if you don't add filtering. This is the reason that fast ADC's usually require some sort of amplifier/buffer stage up front to make sure the sampling cap can be charged fast enough to meet the frequency requirements. Cheers Rob
Thank you for your additional information. I wanted to keep the (tiny) resistors on the Wemos. This is why I ended up with such high values. But the current should be the same as in the original Wemos(@30 volts) .
@@AndreasSpiess The issue is that when the ADC sampling capacitor is switched in at the start of the sampling period it is discharged. The mid point of the divider will drop to 0volts. The cap then charges through the 2meg resistor. Even though the cap is only a few pF it still has a time constant. If the sampling time is short enough the cap will be sampled before it is fully charged and the reading will be off. A big capacitor at the junction of the divider can provide the charge to charge up the ADC cap quickly without significant drop in the voltage at that point. Cheers Rob
Don't forget about adc noise reduction in Atmega. Automatic sampling, averaging without CPU intervention or built-in comparators & other advanced features in bigger Atmegas, STM32, nRf52, and most other arm uCs.
Great video. Just for feedback, the first adc is called flash adc, flash for lightning speed. Another idea, adc chips are good for generating a binary value for raw signal within specifications, but the limitation often lies within the controller to manipulate the binary value to usable data. Often, the dilemma is within the controller's ability to manipulate double float values. Some controllers can only handle 8 byte floats. So calculating a thermocouple voltage becomes an inaccurate reading because of the float calculation errors. Not just thermocouples, even ntc resistors. So usually for such operations a good math library is needed to overcome these hurdles but capture rate of course would be at a disadvantage.
@@AndreasSpiess i'm using the 8266, moved all my heavy load to the STM32F103C8 due to power requirements. I like these ESP devices; however, I'm interested in another solution where they don't offer a flash.
The ADS1115 is a good choice, BUT with my ESP-32 I had some trouble due to the fact that they was in the same casing and WiFi was enabled/used. The results was overlayed with lots of noise. The method of averageing takes to much time for my project. I need the result ASAP and the ADS1115 is anyway no the fastest. But I was able to solve the issue and I want to share this with you guys: First step was to build (3D Print) to boxes which fit in each other with a small gap. The outside of the small box got covered with a copper film. Afterwards I fit the small box in the big box. Why two boxes? Quit simple : isolation. The copper film I connected to ground. Inside the box I placed the ADS1115. For the cables I made small holes and the cable which is going to the probe is an coaxial. Problem solved. What I want to say by this is, that the interference of the RF signal has not to be underestimated!
Very nice video. One consideration though. The serial ADC will always need the same time independent of the measures voltage. As the comparator can only output "too high" or "too low" it can not detect that it hit the right voltage and finish early. It will always increment to the final / least significant bit. It can however be accelerated if you choose to reduce accuracy.
I have a similar adc module with just 2 16bit ADCs and 50Mhz sample rate. It's like 40usd on eBay. It would be incredible if you want to make a video about how to read data off these cracy fast parallel lanes. My primary goal was to reduce cost. I thought it would be interesting if we can dump all the ADC data straight into a ram with some hard logic, and read the ram with some cheap slow microcenter. Thus eliminating the need of fpga.
Great timing. I'm building a flash joule heater, and need to measure the voltage of the supercapacitor before and after the flash... i was just going to use an arduino and not even think about it - because I didn't want to do the research. Thanks much! Edit: after watching, an ADS1115 or equiv is what I need.
Yeah... thought this is a bit less effort. Welcome to reality :) . These 2.7v 500F rh2ings don't want to release their energy instantly. Damn physics...
Nice video!. When you are in the presence of glitches, one way to improve accuracy is to take a lot of measurements and calculate average. Then compare all samples with the calculated average. Take out all samples outside certain margin. Let say 10%. Then calculate the average again. You can continue to get better results. Also you can use statistic formulas as standard deviation to decide when your result is accurate enough.
I agree you can reduce noise with those methods. Non-linearities will stay and unfortunately, the speed will decrease. But often, this does not matter.
Excellent video, thank you! A related interesting subject for a future video might be how to safely and accurately measure the voltages of a multi-cell battery. After a couple of evenings of googling I have found out that's not trivial.
Maybe thats what you are looking for lazarduinoads1115.blogspot.com/2020/03/voltmetre-de-precision-avec-arduino.html I have a 4 cells version on trial now facebook.com/photo.php?fbid=10220134681951223&set=pcb.1665059790308923&type=3&theater
The typical manner is to have the most negative connection coupled to ground of your circuit, and then have voltage dividers for each cell. For a 4 cell battery, you could either use four identical voltage dividers, or preferably have four different ones each tuned to voltage ranges for each of the series cells. For simplicity sake, if it were a 4 cell lithium battery, 5v, 10v, 15v and 20v would be good voltage divider ranges to use for monitoring each successive series cell and being the most accurate for each in turn. Being able to measure the individual cells would be nice, but as you suggest, its not trivial, as you would need to somehow keep changing the ground/negative 'reference' when measuring each cell.
@@matzke089 That could work, yes, but has disadvantages: 1. Relatively expensive, especially for small batteries. 2. High power consumption, especially for small batteries. 3. Communication between the ESPs will have to be engineered carefully. 4. Generally complex and not elegant. A "brute force" solution, if you will. I like elegance and simplicity (the two usually go together).
at 9:30 describing SAR, you first say it would take 4000 samples to get the result, this is not SAR. SAR only ever takes the number of clock that = number of bits so a 12 bit ADC SAR would take 12 clocks. BTW, I do love your videos and find them very well presented and prepared, I am just trying to ad to the excelent presentation you have here for new engineers that may miss interpret the info
a quick suggestion: can you plz do a video about "more in-depth" of voltage dividers using resistors, trying to understand more about voltage dividers and how the work. thank-you and keep up the great content
Just got to the end, you covered off some of my earlier comments.. nice :), i will leave them there to hopeful help noobs to the ADC world and as a text reminder. Awesome video keep up the good work
Hi , about Arduino, i know that you know but is to clarify to others. If you use 5v in your formula and change supply voltage, it is clear that it will not take good values I usually work with the internal secret method for measure Vcc (you don't need any other analog pin for that). Also you can use low noise ADC sleep mode to measure.
Most modern high-speed ADCs are exclusively pipelined architecture ADC. This architecture does include a lower resolution flash ADCs but it does not need all the extra comparators to make the 12-bits. It gets 12 bits by staging the quantization. Also, the conversion does take a few cycles to compete when compared to a pure flash ADC.
@@AndreasSpiess You are exactly right about not being able to do anything serious with the on-chip ADCs. Many people have this misconception. I enjoyed your video.
Danke, Andreas, du bestätigst meine Beobachtungen. Ich verwende auch nur noch die kleinen, externen ADCs. Ein Hinweis noch, die Adafruit-Treiber für ADS1015/ADS1115 könnten beim ADS1115 Probleme machen. Der ADS1015 (12Bit ADC) arbeitet perfekt. Für den ADS1115 ggf einen Fork der Adafruit-Lib von User soligen2010 nutzen.
@Jan. True. I just used this simple one, also to show the formula. Many people store all values and then create the average, which is not good in a MCU environment.
I boosted the ADC sample rate of an Arduino UNO at around 30800 SPS for audio purposes, the sound wasn't really great but when I discovered I can use the 1.1V reference, it immediately made it way more accurate! The issue was that the input audio signal was so low it could only be heard in a single bit.
I recently built up an electronics test setup that utilizes a bunch of sensors in order to create essentially a DIY motor dynomometer. I've yet to finalize the mechanical design but I am using an ADS1115 in order to read an ACS758 current sensor and a voltage divider so I can record and graph the current over time as fast as possible while the motor spins. RPM is read with a cheap light gate style encoder and a 3D printed encoder wheel, so I can control the CPM. Some motors might be geared down to a few hundred RPM, some brushless motors in excess of 30,000RPM. I'm reading currents in excess of 75 amps at peak which is why I'm using a hall sensor and I also want it to work for much smaller motors too, so I wanted a resolution of ~0.05 amps, hence using the external 16 bit ADC. One problem I ran into is sensor drift. I noticed by just sitting there doing nothing over the course of a couple minutes the ADC read values can drive by up to 50. Other than calibrating to 0 at the start ever each test (which is my current workaround) is there any way to try and stabilize this?
Grutti Andreas! In your example at 3:57 , you do not strictly need to apply a clock to the comparators, because only one comparator-value shall change at any given time. What can misfire is the digital truth-table afterwards, (as multiple bits shall change, e.g. going from 1 (binary 01) to 2 (binary10)) - A Gray-code encoding mitigates this too. If the clock can be avoided, the flash-ADC will be even faster.
The Arduino boards *do* provide a pin for the reference Voltage, it's clearly labeled as AREF. The libraries make it easy to use and many beginner tutorials mention its existence.
Hello Andreas! I found your video when searching about adc because i am trying to trick a STM8S003F3 without success. The STM8S seems to be running at 3,3v and has a 10k linear pot connected to 3,3v VCAP, GND and the variable output to PD3 , which is described as AIN4 and ADC_ETR. It stands for External trigger. From what i understand, the ADC external triggering means it is possible that the ADC only reads on some programmed event. Turns out that when i rotate the POT, even when reading with a digital meter, the voltage values are not linear at all and behaves differently along the POT range. It can read steadily on the starting range, mid range it will fluctuate values and even read 0 and on the end range read some steady values again, higher than on the start range. But the device works correctly. What can be the cause for this? The STM8 adc pin can be programmed to somehow behave like this? Or expecting some pwm signal synchronized with its clock? I don’t have a scope to analyse this in a better way. If i input the POT output to a wemos d1 8266 it will read accordingly to the digital meter, but i cannot put the wemos bypassing that signal to a pwm output back to the STM8, it will not work the same way as if the POT is directly connected to the STM8. Do you have a clue about why this happens? Thank you
@@AndreasSpiess ok. Thanks anyways. Maybe it would be something good to figure out. I gave up on that approach because i don’t own a scope but i really would like to know what’s going on behind the scenes. Cheers!
Thanks for another fantastic video. Just for curiosity, if I insert a delay(1) between each analogRead(A0), the measurements related to WeMos D1 Mini have more fluctuations than without it. Is this due to the capacitor in the Sample and Hold circuit that discharges faster by inserting delay(1)? It is evident within low serial baud rate. Thanks
I too had major issues with the ESP32 ADC. Drove me insane. I'd been using the UNO prior to that with 1.1v reference and was very disappointed with the ESP32 ADC performance. Bullshit indeed lol! Changed to an ADS1115 and very happy. So happy I bought lots of them and won't use the built in ADC now. If you're going to use an ADC, might as well use an accurate one.
Thanks for the video, I had a problem with the adc fluctuating reading with my arduino, I never knew that there are imperfections in these microchips. I wish you also compared an STM32 board like the black pill (STM32f411)
Here you go, here is the code to help get the ESP32 ADC to behave better: // configure ADC for getting battery voltage pinMode(VoltageSense, INPUT); // Battery Voltage Sense input analogSetPinAttenuation(VoltageSense, ADC_0db); // set the attenuator for the ADC to allow 1.1V input range analogSetWidth(11); // set the resolution width to be 2048 instead of the default 4096 to try and reduce noise analogReadResolution(11); // set the resolution to be 2048 instead of the default 4096 to try and reduce noise
Also another trick I found is to add a delay directly before the read, it helps get a more stable reading too! 50ms worked well, but you can’t always do that obviously.
Great timing! I just got ten sliding potentiometers and was wondering how to measure their values. Found another video that shows how by connecting all the value pins with a diode to a single analog input, the VCC pin on the potentiometers can be muxed using digital outputs one after another. Also thinking of using a GPIO extender chip like the PCF8575 for this, so I could measure about 16 analog values (slowly) using a single analog pin in the Arduino with pins left over. Will be a MIDI controller project, don’t need resolution or speed or accuracy all that much imho.
Another problem with multi-channel ADCs on most microcontrollers is that the samples are not taken simultaneously, but interleaved. Better ADC and MUX combinations are possible, that use sample and hold before the mux, so even if the ADC part is doing interleaving, the values being produced correspond to the same moment in time for all channels.
@@AndreasSpiess But two of the ADCs you linked to, ADS1115 and AD7705, are sigma-delta. They're quite common, especially if you're looking for cheap breakout boards.
Me at 16 mins: "ah... bugger" :D Had some "extensibility" ideas in mind for a project to potentially later hook some ADC2 pins up to some inputs... Still, good to know an I2C board is an option if I later want to do go that route and still use Wifi. Video wins just for that, but it had already won several times over for teaching me about how ADCs work!
I am wondering why the chip designer for ESP32 didn’t add shield to the ADC part so that the result will not be influenced by the EM coupling. Over sampling with noise shaping can increase the precision as well as remove the glitches, which was what I was expecting this video to include. But this method is really a simple and convenient way for the purpose, which is what I haven’t thought of.
Remember that in Arduino Uno and all the Arduinos with ATMEGA328 you can select an internal stabilized voltage reference, to avoid the errors with power supply fluctuations
VRef is applied to the inverting inputs ( - ) of the operational amplifiers. When an input signal (voltage) is applied to Vin (non Inverting Inputs) ( + ), then the operational amplifiers will turn on accordingly because of the rules of the operational amplifier logic as a comparator. The way you have it, if Vin is zero, then all outputs on the operational amplifier is logic HIGH. So LEDs on the outputs on the Operational Amplifiers would turn off instead of turning on when an input signal is applied to Vin.
Notf sure overall, but the esp32s2 has a bit width of 13, compared to the esp32's 12 - which is significant. I don't think it's shown in the documentation yet? However idf alerts you to this when compiling and it can be seen in the respective header files (of course it works too!).
@@Sovvyy It's just a marketing. I've tried this to monitor charge of the battery -- it's useless in the same way as regular ESP32 or ESP8266 that ESP32s2 should replace. Use only external battery controllers chips, internal ESP ADC is useless for such sensitive things. And what I hate the most with ESP -- this "filtering" formulas with magic numbers -- you need to adjust it for each and every chip. I never had any 2 chips in ESP32 or ESP8266 lines with the same ADC errors.
Nice video, I always interest in electrical measurement sensor. However I've noticed you use 2^n-1 for your calculation (eg. resolution, voltage conversion), it's always caught my attention because there 2 side for this all over the internet, whether use 2^n or 2^n-1. To this day I'm used 2^n for my calculation, LSB (resolution) is given by Vref/2^n and voltage sensing using value*Vref/2^n, it's because my understanding that there only 2^n step (value begin at zero) so the maximum value of sensed voltage (FS) is always 2^n-1 sized, not equal or exceed the Vref itself. If the calculation related to FS then I will use 2^n-1 so LSB=FS/2^n-1 because FS=Vref-1LSB. There so many tutorial, for example 10bit built-in Arduino ADC, using 1023 or 1024. I think it's interesting topic to discuss this fundamental math for ADC.
FYI. Re: ADS1115 It seems to be OK to supply input voltage to the ADS1115 even when it is not supplied with Vcc. i.e I have a project where I am measuring the voltage of a car battery, using a ADS1115 and a ESP32 , using a voltage divider to reduce max of 16V to 3.3V on the input of the ADS1115 If I remove the supply voltage, the ADS1115 seems to survive even though the input was 3.2V and Vcc was 0V. Possibly this is because I'm using 4.7k in series as part of the voltage divider, so the current would be limited to about 3.5mA. In hindsight, I will try using higher values for the voltage divider and use a reservoir capacitor to reduce the effects of the higher impedance input. e.g. try 47k and 12k as the divider, so that the max current would be only 0.3mA PS. I just accidently destroyed on channel of ADS1115 by shorting the 4.7k resistor with a solder bridge, hence supplying 16 to one input, but the other inputs still work OK. So when I rebuild my prototype I will change the voltage divider resistances.
So you were lucky that only one input was destroyed... I try to stay inside the limits of the data sheet. Maybe the chip has an input protection diode towards VCC and as you write, this protects the chip during power off. And you are right, you can use higher resistors for the voltage divider.
@@AndreasSpiess I actually destroyed 2 inputs, because I shorted 2 resistors. However the other 2 inputs still work OK. Before I replace the module I will do some more testing on the 2 remaining operational inputs to see how much current is taken by the ADC inputs when the battery I am measuring is still connected, but the supply to the ADS1115 is removed, i.e I don't power my entire sensing board. Like you say, there is probably a protection diode to Vcc on the ADC inputs, but this diode is probably only designed to protect when the supply is removed from the ADS1115 when there is a series resistor connected to the ADC input. I wonder if someone had a 1.5V battery connected directly to the input of the ADC, whether the input would be destroyed if the ADC was not powered. I also need to experiment with higher value resistors in the voltage divider, to increase protection, The manufacturer seems to recommend resistor dividers with values around 1k Ohm in total, to prevent measurement error, according to a post on their forum. However the input impedance of the ADC is 3M Ohm, so higher values should not cause significant measurement error and adding a reservoir capacitor will also compensate for the higher resistor values albeit slowing response time.
Bottom line is that the accuracy actually required for a given application drives the decision of if built-in MC's ADC or external ADC will be required. The ADS1115 is my go to solution if it's required, very flexible and cheap.
This was super helpful and timely for me. I've been struggling with how inaccurate the ADC is in the ESP32 trying to get good voltage readings. I'm assuming that I don't need a voltage divider if use the ADS1115 to measure an 18650 Lion and if I power it from the ESP32 it wont cause a current draw when the ESP32 is in sleep mode? I have learned some useful stuff struggling through this process though, like how to switch on and off a voltage divider using a N and P Channel MOSFET Module (CJMCU-4599 Si4599) so that the divider doesn't constantly drain the battery (thanks Ralph S Bacon). I also looked into using the INA219 board, but the .1 ohm resister creates a constant draw a battery as well. I'm going to give the ADS1115 board a try, I just ordered a few from Amazon. Andreas, thank you so much for presenting this information in such a clear and concise manner. Cheers!
Pay attention with the ADS1115: It cannot measure much above its VCC. So if you connect it to 3.3 you cannot measure 4.2 volts without voltage divider.
If you made the resistor ladder uneven, and stuck with 4 comparators, you could get useful levels e.g. 3.6 3.8 4.0 4.2v dedicated for monitoring lithium cells
The ADS1015 is a drop-in replacement for the ADS1115 and can provide much better stability and speed, but is only 12 bits. Conversion is 3 to 8 times faster when reading values over I2C and sample rate is more than 4 times faster. It's also half the cost and has the same footprint.
It's also possible to use a smoothing capacitor on analog values instead of doing too much work in software. I wanted to use a 100K ohm pot going from ground to Vcc and there's not enough current to overcome the internal leakage and you get bad results. Corrected with a 0.1uF capacitor on the pot slider and it removes the error on those internal ADs and you still get the advantage of low current consumption through the 100K pot.
@@AndreasSpiess true. I only use the internal A2Ds for basic applications. I just begrudge wasting a lot of current through a pot connected across the power rails. The capacitor method works really well.
The external ADC at 2.5V didn't show 2.4994 as you say as per the reference but 2.4594 at 22:22 in the video.
You are right. My mistake!
ua-cam.com/video/AIL36S6NpDA/v-deo.html awesone project
yup. I felt a little spoofed.
With that inaccuracy it seems that an external ADC isn't really that much better than a software-compensated ESP32 ADC.
However, the precision of an external ADC could be interesting for scaled up projects: you are trading the extra costs for an ADC to skip the individual calibration process of each ESP32 ADC.
@@tigerresearch2665 The external ADC only performs so bad due to his own "correction formula" that very likely only corrects errors on the analog end (aka from V_ref to the PGA inside the ADC). So it shows the correct value at this one point, but not anywhere else. One error that affects the analog end, which I would suspect here: High sampling rate -> low input impedance -> voltage drop Correcting ADC values in code like he does is not something anyone should do. The first step is to correct the hardware/setup!
Let me compare that to a car that constantly steers to the left: Do you just "counter steer" to make it straight... or do you fix your car so it drives straight?
Andreas Spiess So, the conclusion of the Video is wrong?
I am using an ESP32 to monitor the battery voltage on my boat and send an email if it gets too low. I found the measurements very noisy due to occasional glitches. My solution was to take 8 readings and ignore the largest and smallest and average the middle 6. I do this twice and average them. It is quite easy to do in the code as you just have to keep track of the largest and smallest measurements and subtract them from the sum and divide by 6. I found this better than longer averages so that I did not get multiple glitches in my sample period. The result is very stable and quite fast. I very much enjoy your videos I am a retired circuit designer. I designed test equipment for HP (Now Keysight) and things like cable modems in the early days. Now it is a hobby.
Good idea because you can eliminate the far glitches. I did not investigate the distribution of the measured values, but maybe they have a pattern you break with your formula.
I like my Keysight MSOX3045 ;-) Good products!
@@AndreasSpiess The distribution was not Gaussian. It was mostly correct values with a few outliers. It is actually a pretty standard way of filtering. There is even a name for it but I can't find it any longer. I learned it from a statistician who worked for me for a short time.
Would it help to put a cap (as low Pass?) between the divider and gnd.
@@michael_the_chef It has been a long time but I think I measured with a dead stable voltage so the answer would be no.
@@L-36 ok thanks, i am planning to use an ads1115 and an esp32 to monitor a battery Bank of 4
Thanks for this video! I've been puzzling over how to reduce the noise of a voltage reading for a project without losing responsiveness introduced by filtering. Based on your video, I switched over to an external ADC (ADS1115) and it solved the problem. Much appreciated!
Glad I could help!
For a very long time I paid my studies by building accurate DAC's for the growing digital industry. Making them precise is a science! 😊
Sounds neat, were they hand-matched R2R ladders? I always wonder why you can’t just buy R2R resistor networks in single packages, like a SIP10. Though these days I’d be tempted to just use a half-bridge and low-pass filter, like a good-old class-d amp.
@@Scrogan Try www.digikey.com/products/en/resistors/resistor-networks-arrays/50?k=resistor+array&k=&pkeyword=resistor+array&sv=0&pv9=392120&sf=0&FV=-8%7C50&quantity=&ColumnSort=0&page=1&pageSize=100
Yup, I have a esp32 based ph/temp/ec meter I'm doing for hydroponics github.com/physiii/grow
@y2ksw1: I can imagine that this is not trivial!
James Barclay Yes, with my never patented fool proof method nobody ever thought about before. Pure science 😊
I never get tired to watch your videos. I like the fact that you put your personal opinion in each fact. I understand better this way. Thank you
Thank you for your nice words!
The AVR ATmega328p used in the Arduino, and other AVR chips, have a special sleep mode for ADC noise reduction. It keeps the ADC running, but stops all the digital processing. See the "ADC Noise Reduction Mode" in the datasheet. It could be that the external ADC chips appear better simply because they are physically further away from the running digital logic.
Interesting. I didn't know Arduino had a sleep mode.
@@Slutuppnu I don't know about the Arduino library, but with AVR-GCC you have fine-grained control over it. www.nongnu.org/avr-libc/user-manual/group__avr__sleep.html
PS: The Arduino IDE already uses AVR-GCC, so you can use the functions that I linked from Arduino projects as well.
@ is it worth it with a 10 bit adc ? (Atmega16A)
@@Slutuppnu Most AVR chips (the ones inside arduinos) don't have just one sleep mode but a few of them. It has (as has been said), an ADC noise reduction in which the CPU and some other digital parts of the chip are turned off in order to remove noise from the ADC signal; it has a CPU sleep mode that allows you to save a bit of power during delays or when waiting for a short timer to wake you up; and finally it has a deep sleep mode, in which the CPU, fast clock signal (either the internal fast oscillator or the external crystal oscillator), and most peripherals are turned off, except for the WatchDog Timer (low power timer) and a few more things like pin interrupt wakeups. This allows for very low power applications, in which you can put the chip to sleep and only use a few microamps of current, which for battery applications is really useful. For those types of applications, you have to take care and desolder all the LEDs on the board that you can't control, make sure to avoid using any LDOs onboard, etc.
The ADS1115 is my favorite ADC for microcontrollers. I started using it when I moved over to 32bit microcontrollers, but needed a way to read 5V automotive sensors (using a Teensy 3.2). I had 4 sensors (an MPX5700AP pressure sensor, an NTC thermistor, an AEM Wideband oxygen sensor and a 0-100psi oil pressure sensor), the ADS1115 had 4 inputs, it was a match! I didn't have to average values to get a stable reading, like you said, the stability of the ADC is fantastic.
Thank you for sharing your experience. This seems to be a perfect solution.
Awesome. Thank you for the great content! Just my two cents:
1. I've had to learn the hard way, that many (all?) of the ADC that have a MUX only have a single S/H aufter the MUX. This leads to problems, when reading signals on different channels with very different voltages in quick succession. There is quite a bit of "crosstalk". It's really important to make sure that the inputs to such an ADC are low impedance.
2. When I do averaging on a uC, I do 2^n samles (8, 16, 32...1024) samples, add the read values up (make sure to prevent overflow) and just use shift for the division at the end. Especially on uC without an FPU this speeds the averaging up quite a bit.
Good info. Thank you! I would have expected only one S/H, though, because it is usually part of the ADC.
@myselfremadedid you ever find an answer?
Nice video.
Remember to consider the input impedance of the ADC, especially if you use high value resistors for a voltage divider in front.
since the input is in parallel to the low-side resistor of the voltage divider only its value really matters - well actually it is of course the parallel-combination of the upper and lower resistance in the voltage divider...
And for the ATmega328 the input resistance of the ADC-channels is astonishingly high! I tested it with discharging a 1µF capacitor from 5 V to zero through a 100 MΩ resistor - it took about 15 minutes to discharge and the resulting curve indicates an input resistance beyond 1 GΩ.
Uwe Zimmermann
I was going to ask what you used to measure that voltage, but I guess you were just using the ADC itself. I imagine most of the leakage current will be through the protection diodes, if it had none then it would be a real recipe for ESD. Which reminds me of a friend using his atmega ADC for measuring back-emf of a solenoid. Directly on the 12V half-bridge output that was powering the solenoid. With no extra diode or resistors. The arduino got really hot, but somehow the circuit actually worked.
@@uwezimmermann5427 I am not talking Any specific microprocesor. I had an adc in a low power circuit where I wanted to messure the battery voltage, here I used a resistor divider in 10 Mohm Range to avoid draning the battery. In this applikation the messured voltage was wrong due to current flowing into the mcu.
If you tweak the adc settings in the ESP32 down to 10 bits, and some other changes, the noise and accuracy gets a LOT better, I’m not at my computer at the moment but I will post the code when I get a chance.
Please leave a link on your comment when you do. Thanks
Augustine Nwafor - I posted the code in the comments, it’s here somewhere.
Here is the code of the @@TheDefpom comment:
"Here you go, here is the code to help get the ESP32 ADC to behave better:
// configure ADC for getting battery voltage
pinMode(VoltageSense, INPUT); // Battery Voltage Sense input
analogSetPinAttenuation(VoltageSense, ADC_0db); // set the attenuator for the ADC to allow 1.1V input range
analogSetWidth(11); // set the resolution width to be 2048 instead of the default 4096 to try and reduce noise
analogReadResolution(11); // set the resolution to be 2048 instead of the default 4096 to try and reduce noise"
@@skuashh cool， will try when i have a chance
Greetings from South Africa Andreas. Well done as usual! I am always surprised how much information you share with great clarity and structure in such a short video clip. When I learnt this stuff (long before internet), choices were fewer and one had to study the data sheets thoroughly (or copy somebody's example from Elektor, EE, ETI etc). Microcontrollers (with built in ADCs) only arrived later and choices were fewer - which was a blessing, because obsolescence took longer. Hobbyists today should be thankful that you share years of experience, crammed into a nutshell...and for free. Love your work and your very structured method of teaching (unlike my ADHD brain which likes to skip around). Best Regards, Allen.
At 12.07, I love the way you dodge the 1023/1024 argument!
I had enough discussions about that in my life. So I once used 1023 and once 65536
@@AndreasSpiessWhy was there ever an argument? (1023/1023 =1) Just curious
@@TheRainHarvester Because of the 0
@@AndreasSpiess if you can measure the voltage at the high rail, you divide by 4095 so you get 1* vcc.
Very interesting video as always, thank you.
Another benefit of using the ADS1115 is that the internal PGA can be configured to read a differential signal, this is very useful for measuring some sensors, although it also results in a slower data acquisition rate.
True!
This trick can also be used in general by the Arduino by using two analog inputs connected to the signal direct and the signal amplified by a fixed op amp.
Hints for those starting out -- for built-in ADCs, throw away the bottom bit as it's most likely useless anyway. You can average samples equal to a power of 2 by adding the integer values and doing a right shift -- for example add up 8 integer ADC samples and right shift 3, a useful trick when your microprocessor doesn't have a divide instruction! What's your signal bandwidth? Adding a single pole RC low pass filter in front of the ADC can help. And do not expect the input impedance of the ADC to be constant.
Good idea. Thanks!
Another great video from a very competent presenter.
Just a point for a possible misconception: there are two other ADC architectures in common use, namely pipelined and sigma-delta. The first tries to keep the number of parallel comparators required in check by converting the output of a comparator stage back to analogue using a DAC, subtracting this output from the original input, and feeding the result to another comparators' stage. Three stages are quite common with typically 4bits per stage; leaving aside the unavoidable latency the conversion takes one clock cycle like the comparators' ADC. This breaks for high resolutions because analog subtractions accumulate too much error. Then sigma-delta comes to the rescue where in fact a control system tries to get its 1 bit output bitstream average as close to the input as possible (its operating principle is not easy to understand in detail though); 24 bits and higher resolutions became possible.
Successive approximation is favoured by the MCU manufacturers for a very simple reason - it requires a single analog comparator and an R-2R DAC - cheap, accurate and reproducible architecture that covers nearly all practical DAC needs. Moreover it can be difficult to find a standalone successive approximation ADC as most are either pipelined (higher sampling rates) or sigma-delta (for high, eg audio, resolution).
Again for a typical maker these are too finer points but I wanted to make sure they are mentioned in the footnote. Thanks again Andreas for your work
Thank you for your explanations. I always learn something from my viewers! As you saw I am more a user and not a specialist in this matter.
How timely. I'm just trying to do this right now to monitor the voltage on a battery in my mailbox alerter. :). Thank you!
For monitoring a battery the internal ADCs should be ok.
Thanks Andreas for this interesting video. I've used ADC lots of times, and never had concerns about the quality of the conversion (noise reduction with average method). In a current project, getting feedback from the servo's potentiometers for a robotic arm, I've used a calibration table, but now I'm going to introduce your formulas. Thanks again, Juan.
It depends on the desired accuracy. I compared, of course, with the data sheet.Calibration tables are good if the ADC is dependable wrong and not stochastically wrong...
Watched already a lot of videos of your channel and at the beginning it was a lot of „häääää, versteh grad überhaupt nix“ but from one video to the next I start to understand what you are explaining us. Have to say that I am a totally new beginner.
My goal is to measure the 2 12V Solar Batteries in the OFF-Grid garden of my parents in law. Tried a lot with the ADC from the ESP32 itself and as u mentioned it’s readings are just bullshit.
Thank you for introducing the ADS1115 to me. I guess that will fix my measuring problem and can finally bring the project into production.
As I am original from Vorarlberg and worked many years in Kanton SG I love to hear your swiss accent as it remembers me of that great time in my life!
Electronic wise you are my personal hero ;-)
For this purpose you probably could also use a INA226 as suggested by other commenters.
And I am glad you learn something with each video. We all started as beginners in this field.
Andreas, you're getting pretty good at this. In a foreign language, even more impressive.
Thanks
He has a Swiss Accent though! :-)
@@jamesgoacher1606 Too late to change that. But look at Big Clive with his Scottish accent.
I vote for the content, not the form!
This is such an important topic. I send a million thanks and encourage this topic.
Getting signals INTO the computer is probably just as important as having the computer to begin with.
This topic makes me consider re-writing all of computer science just to make an analog computer.
You find analog computers if you google. They were used befor the cheap silicon
23:35 Really freaky video compression artifact at 23:35 -- the RF connectors appear to randomly telescope in and out!
Because they are not soldered in, he just put them on the PCB just to show the board.
It's not a video artifact.
@@shreyaskul Yes, on closer viewing, I think you're right. They seem to be rocking back and forth.
Shreyas is right!
@@AndreasSpiess Hey! Thanks for your comment.
Can you share which IC you have used in your ADC board? That would be great :)
Hi Andreas! Great video as always! One note about the ADS1115 is that it is actually a 15-bit signed differential ADC( AC !!! ), or 16-bit single-ended. Differential
measurements offer more immunity from electromagnetic noise. This is useful when using long signal wires or operating in an electrically noisy environment. I have been using an ADS1115 to measure the amperage of our utility inlet with a sct-013. The fine art of accuracy I've seen is matching the sampling rate of the adc to the frequency of your local provider to sample the highs and lows of the sine wave as accurately as possible (50Hz here in ZA).
You are right with differential measurements.
Concerning 50Hz: You have to synchronize because the ADS1115 is slow. A diode and a RC filter probably would help to get DC out of AC
@@AndreasSpiess on the sct013 each 10A represents 1v. Thus getting Amperage readings are easy and safe with non invasive clamps. Only when it comes to the measurement of voltage, territory gets unsafe. I have opted for a 220v primary, 12v secondary to convert ac to DC proportionally followed with a voltage divider and capacitor. With both voltage and amperage the apparent usage can be calculated to identify power consumption patterns at home. I cannot thank you enough for the extensive and concise research you have given to this community! You inspire us!
You have confirmed my discovery. Thanks! Unfortunate that we have to rely on "small and casual" posts or trial&error methods to correctly use the esp32.
Fortunately, we have Google and the internet to share our learnings ;-)
Thanks for this awesome intro to the ADCs!
I'm going to measure voltage of my LiPo cell, so I know when to charge it :)
I do this too for most of my battery operated nodes
Thank you for the breadth of the experiments to help reinforce the lessons. Seeing how externally sourced ~5v can impact the readings was an "of course!" moment, since I'm used to using a good regulator versus an adjustable power supply for projects. Spotting things like how the radios on the chip may affect the ADC are very good clues. I'll investigate some of the external ADCs, since I'm thinking about a project that will use a mesh radio network and I had planned to use analog inputs as well... thanks again!
Some distance to radios is never a bad thing ;-)
I like the fact that your opinions/observations are unfiltered. You just tell it like it is. "Subscribed"
Welcome aboard the channel!
I'm amazed every time I see your great videos!!! You are testing stuff that went true testing before manufacture, but you made it better - why isn't Arduino listening??? If Arduino and ESP's had better ADC, the cost would not be that much higher - right? But now we can use your recipe and suggestions! Thank you again Andreas. And wishing your finger to heal FASTER!
I am sure they have their constrraints. Create chips for a few cents with 3 CPUs and Wi-Fi is for sure not easy.
Thanks for putting out such high-quality videos regularly!
Glad you like them!
Perfect! Thank you for this video! I am working on a project now with analog sensors and i learned everything I needed to!
Glad it helped!
That was very enlightening, too bad my favorite MC ESP32 yielded such weak (lab) results - still plenty accurate though for many projects. Will look into the ADS1115. Dankeschön !
I also use the internal ADC. If you know what you can expect, it is ok.
Reminds me of school and r2r ladder dacs. I get my fix of andreas every sunday. Lovely stuff.
Glad you like the "stuff" !
Great video. One thing to note. If you use an external i2c ADC, you can not set up any sort of interrupt driving timing to collect sames in the background at a given rate since i2c does not work within a ISR. This leads to a bit of work to make sure loop() gets called at a reasonable rate.... Just something to be aware of.
Good to know. Maybe you try a second RTOS task on the ESP32
@@AndreasSpiess This was on the AVR architecture
Neat video! I’d have liked it more if you went over the details of successive approximation registers, and how a delta-sigma ADC (or DAC) works, but maybe that’s outside the scope of this video. On the topic of ADCs and DACs, I’d be interested in a video on DSP algorithms being implemented in an arduino, be it for audio or for filtering out narrow-band noise.
Also when you said that SDRs use these very high speed GHz ADCs, I don’t think they actually do. They use lower speed ADCs with mixers and tuneable local oscillators to generate the I and Q signals. Direct sampling SDRs are very rare.
I’d also have mentioned the fact that the ADCs have a finite input impedance, so putting a network of resistors in front of one without buffering it after could cause some systematic error. Well, you often use an instrumentation amplifier there anyhow so maybe it’s not too useful of a tip.
1. This channels is for makers, so the most important decision is what not to show.
2. I define SDR not only as "SDR receivers for the general public". I define them as "digitalize the signal and do the processing in software". So radar and many other applications fit my definition.
3. You are right with the input resistance. If you calibrate the readings I assume you can live witout an instrumentation amplifier for 10 or 12 bit resolution. For more precise applications you are right, of course
I recommend using median filter for removing ESP32 ADC glitches, averaging is not good for this kind of noise
The formula @Andreas Spiess is using is not the usual average formula. He uses a so called filter with feed back and I don't know why he is calling it average but that one for sure is not just an average. And the second point when he says "when I use 10 samples or even 100 samples in that formula" .. those are not the actual samples. That number tells how much we should fed back to the initial value so basically to explain it we have an initial x=0.551 and in the loop he is calculating the next x by taking 99 parts of original x value (0.551) and adding 1 part of the new analogRead value thus it basically means we are just drifting to the left or right of the initial value and the amount of drift is expressed in the ratio we are using 1/10, 1/100, 1/1000 and so on parts of the new value mixed back to the initial one. Also this number will be the speed of conversion to the initial value so the higher value we use the slower is the speed and the more analog reads we need to go back to normal and at the same time the less is the chance to affect the final number by the new erroneous reading. Maybe @Andreas Spiess does not fully understand the formula he is using which I doubt but who knows maybe he can explain it in some further videos.
@@valcob6617 This is a simple IIR filter (meaning the response never reaches steady state theoretically), its good for quick & dirty filtering
@@valcob6617 Also the filter is prone to accumulating round off error for high filtering coefficients and my guess is this is why he gets 10% error on ESP32's ADC
I had to chuckle a little about the example of measuring a 24V Solar panel and then setting the range to 30V (around 13:23 in your video). A solar panel that you would use with 24V batteries in a real-world scenario would normally be a 72-Cell panel, which typically has an open circuit voltage around 45V @25ºC, and quite a bit higher when it gets really cold. I realise it was just an example in your video, but for anyone wanting to really measure a "24V Solar Panel", maybe increase that range in your resistor divider a little :-) As usual, thanks for an interesting Video Andreas - toll!
Thanks for another great video! In researching my gas sensor project, I came across a method people say improves the accuracy of the readings, though I don't really understand it. You apply this calculation to the reading on the ADC pin: volts = -0.000000000000016 * pow(reading,4) + 0.000000000118171 * pow(reading,3)- 0.000000301211691 * pow(reading,2)+ 0.001109019271794 * reading + 0.034143524634089; (Also note that there is a more efficient way to do this calculation (10x faster), but it yields a slightly different result, so I'm leaving the original one here.)
For some reason the calculation got scrambled when I pasted it in to the comment - apparently having a "+" after a "-" gets interpreted as strike thru... Hopefully you can figure it out.
Unfortunately I am no mathematician. Maybe another viewer can help out.
@@AndreasSpiess I forget how I originally found the calculation, but if you Google "ESP32 ADC Polynomial" you will find a lot of people use it to make the values returned by the ADC more accurate. You noted that the ESP32 wasn't accurate, but apparently it can be improved with this technique. I'm happy with the results I'm getting, but my application doesn't need it to be extremely accurate.
Nailed it again. Thank u for your work. You always have what I need. Explained in full. Perfect
Thank you for your kind words!
Nice explanation on the ADC basics. This can be confusing for beginners and experienced people alike.
The resistor values used when measuring higher voltages(30V) are way to high. With a few meg ohms the ADC's sampling capacitor cant fully charge and your results will be off. As the voltage drops the problem will get worse. You need to either increase the sampling time (not rate) or put a big cap (10-100nf should do) at the junction of the voltage divider. The capacitor will then supply the required charge to charge up the ADC's sampling capacitor. This will obviously mean the sampling RATE will be dependent on the RC values of the voltage divider and the cap.
If you do the same experiment with resistor values in the range of 10 - 100K as the voltage divider , you will get much better results. This is not an issue with the ESP32 , the same will hold true for all ADC's. I use the STM32 a lot and have seen this exact issue. If you have to sample a high impedance source you will have to increase your sampling time. You will also see more noise if you don't add filtering.
This is the reason that fast ADC's usually require some sort of amplifier/buffer stage up front to make sure the sampling cap can be charged fast enough to meet the frequency requirements.
Cheers
Rob
Thank you for your additional information. I wanted to keep the (tiny) resistors on the Wemos. This is why I ended up with such high values. But the current should be the same as in the original Wemos(@30 volts) .
@@AndreasSpiess
The issue is that when the ADC sampling capacitor is switched in at the start of the sampling period it is discharged. The mid point of the divider will drop to 0volts. The cap then charges through the 2meg resistor. Even though the cap is only a few pF it still has a time constant. If the sampling time is short enough the cap will be sampled before it is fully charged and the reading will be off. A big capacitor at the junction of the divider can provide the charge to charge up the ADC cap quickly without significant drop in the voltage at that point.
Cheers
Rob
very nice info there Andreas, thanks, it will come in handy...
id be interested in seeing an RTL-SDR ADC compaired to these MCU ADCs
It is quite different. If I remember it has only 8 bits. But of course, it is faster.
Don't forget about adc noise reduction in Atmega. Automatic sampling, averaging without CPU intervention or built-in comparators & other advanced features in bigger Atmegas, STM32, nRf52, and most other arm uCs.
Good to know. Thanks!
Great video. Just for feedback, the first adc is called flash adc, flash for lightning speed. Another idea, adc chips are good for generating a binary value for raw signal within specifications, but the limitation often lies within the controller to manipulate the binary value to usable data. Often, the dilemma is within the controller's ability to manipulate double float values. Some controllers can only handle 8 byte floats. So calculating a thermocouple voltage becomes an inaccurate reading because of the float calculation errors. Not just thermocouples, even ntc resistors. So usually for such operations a good math library is needed to overcome these hurdles but capture rate of course would be at a disadvantage.
True. Fortunately, The ESP32 has a floating point engine.
@@AndreasSpiess i'm using the 8266, moved all my heavy load to the STM32F103C8 due to power requirements. I like these ESP devices; however, I'm interested in another solution where they don't offer a flash.
The ADS1115 is a good choice, BUT with my ESP-32 I had some trouble due to the fact that they was in the same casing and WiFi was enabled/used. The results was overlayed with lots of noise. The method of averageing takes to much time for my project. I need the result ASAP and the ADS1115 is anyway no the fastest. But I was able to solve the issue and I want to share this with you guys: First step was to build (3D Print) to boxes which fit in each other with a small gap. The outside of the small box got covered with a copper film. Afterwards I fit the small box in the big box. Why two boxes? Quit simple : isolation. The copper film I connected to ground. Inside the box I placed the ADS1115. For the cables I made small holes and the cable which is going to the probe is an coaxial. Problem solved. What I want to say by this is, that the interference of the RF signal has not to be underestimated!
Very nice video. One consideration though. The serial ADC will always need the same time independent of the measures voltage.
As the comparator can only output "too high" or "too low" it can not detect that it hit the right voltage and finish early. It will always increment to the final / least significant bit.
It can however be accelerated if you choose to reduce accuracy.
Thank you for the additional information!
I have a similar adc module with just 2 16bit ADCs and 50Mhz sample rate. It's like 40usd on eBay. It would be incredible if you want to make a video about how to read data off these cracy fast parallel lanes. My primary goal was to reduce cost. I thought it would be interesting if we can dump all the ADC data straight into a ram with some hard logic, and read the ram with some cheap slow microcenter. Thus eliminating the need of fpga.
Great timing. I'm building a flash joule heater, and need to measure the voltage of the supercapacitor before and after the flash... i was just going to use an arduino and not even think about it - because I didn't want to do the research.
Thanks much!
Edit: after watching, an ADS1115 or equiv is what I need.
Hopefully it will work out for you!
Yeah... thought this is a bit less effort.
Welcome to reality :) .
These 2.7v 500F rh2ings don't want to release their energy instantly.
Damn physics...
Nice video!. When you are in the presence of glitches, one way to improve accuracy is to take a lot of measurements and calculate average. Then compare all samples with the calculated average. Take out all samples outside certain margin. Let say 10%. Then calculate the average again. You can continue to get better results. Also you can use statistic formulas as standard deviation to decide when your result is accurate enough.
I agree you can reduce noise with those methods. Non-linearities will stay and unfortunately, the speed will decrease. But often, this does not matter.
I didn't know the truth and tricks of ESP8266/ESP32. Another high quality video. Thank you!
You are welcome!
Excellent video, thank you!
A related interesting subject for a future video might be how to safely and accurately measure the voltages of a multi-cell battery. After a couple of evenings of googling I have found out that's not trivial.
Thank you for the suggestioon.
Maybe thats what you are looking for lazarduinoads1115.blogspot.com/2020/03/voltmetre-de-precision-avec-arduino.html I have a 4 cells version on trial now facebook.com/photo.php?fbid=10220134681951223&set=pcb.1665059790308923&type=3&theater
The typical manner is to have the most negative connection coupled to ground of your circuit, and then have voltage dividers for each cell. For a 4 cell battery, you could either use four identical voltage dividers, or preferably have four different ones each tuned to voltage ranges for each of the series cells. For simplicity sake, if it were a 4 cell lithium battery, 5v, 10v, 15v and 20v would be good voltage divider ranges to use for monitoring each successive series cell and being the most accurate for each in turn. Being able to measure the individual cells would be nice, but as you suggest, its not trivial, as you would need to somehow keep changing the ground/negative 'reference' when measuring each cell.
I thought it‘s really simple if you use an individual esp8266 for each cell. It‘s even scalable. What is wrong about that?
@@matzke089 That could work, yes, but has disadvantages:
1. Relatively expensive, especially for small batteries.
2. High power consumption, especially for small batteries.
3. Communication between the ESPs will have to be engineered carefully.
4. Generally complex and not elegant. A "brute force" solution, if you will. I like elegance and simplicity (the two usually go together).
at 9:30 describing SAR, you first say it would take 4000 samples to get the result, this is not SAR. SAR only ever takes the number of clock that = number of bits so a 12 bit ADC SAR would take 12 clocks. BTW, I do love your videos and find them very well presented and prepared, I am just trying to ad to the excelent presentation you have here for new engineers that may miss interpret the info
You are right with the number of 12. There I made a mistake by saying "around 10". The 4000 points was the explanation without SAR.
Están increíbles todos los detalles que enseñó profesor... 👌😎
Gracias!
a quick suggestion: can you plz do a video about "more in-depth" of voltage dividers using resistors, trying to understand more about voltage dividers and how the work. thank-you and keep up the great content
This is too basic. Just google and you will find a lot of stuff.
Just got to the end, you covered off some of my earlier comments.. nice :), i will leave them there to hopeful help noobs to the ADC world and as a text reminder. Awesome video keep up the good work
Cool, thanks
Thanks Andreas, once again learned alot!
Glad to hear that! Thank you.
I was just about to buy an ADS1115, good timing!
I hope you will like it!
Hi , about Arduino, i know that you know but is to clarify to others. If you use 5v in your formula and change supply voltage, it is clear that it will not take good values
I usually work with the internal secret method for measure Vcc (you don't need any other analog pin for that). Also you can use low noise ADC sleep mode to measure.
This is what I showed in video #10 ;-)
I spent much time working in Switzerland, but never heard an accent like yours ;)
So it seems I am special :-)
Great video, I just bought some external voltage reference modules so great timing
Perfect!
I think sigma delta ADC was also worth mentioning which is also used in ADS1115, maybe in future videos
You are probably right. But the video was already very long :-(
Most modern high-speed ADCs are exclusively pipelined architecture ADC. This architecture does include a lower resolution flash ADCs but it does not need all the extra comparators to make the 12-bits. It gets 12 bits by staging the quantization. Also, the conversion does take a few cycles to compete when compared to a pure flash ADC.
You are right.
@@AndreasSpiess You are exactly right about not being able to do anything serious with the on-chip ADCs. Many people have this misconception. I enjoyed your video.
Another great tutorial. I enjoy my time watching your videos. Thanks for making such valuable content. 👍
Glad you enjoy it!
Danke, Andreas, du bestätigst meine Beobachtungen. Ich verwende auch nur noch die kleinen, externen ADCs. Ein Hinweis noch, die Adafruit-Treiber für ADS1015/ADS1115 könnten beim ADS1115 Probleme machen. Der ADS1015 (12Bit ADC) arbeitet perfekt. Für den ADS1115 ggf einen Fork der Adafruit-Lib von User soligen2010 nutzen.
Danke für den Tip!
As always a verry interesting video.
By the way it is also possible to do other things than averaging. E.g. dropping the spikes / outliers.
To remove this kind of "shot noise", a median or percentile filter is a common and easy-to-implement method!
Or simply use "Olympic average"
@Jan. True. I just used this simple one, also to show the formula. Many people store all values and then create the average, which is not good in a MCU environment.
@@AndreasSpiess
Can you plz explain how the averaging formula works ? And how you get the 3.09 number for correction you used for esp32 test
hey man, your secrets are the best! thanks for all this information . greeting from argentina amigo!
Glad you like them!
I boosted the ADC sample rate of an Arduino UNO at around 30800 SPS for audio purposes, the sound wasn't really great but when I discovered I can use the 1.1V reference, it immediately made it way more accurate! The issue was that the input audio signal was so low it could only be heard in a single bit.
You are right. The usage of the internal reference is shown in video #10
I recently built up an electronics test setup that utilizes a bunch of sensors in order to create essentially a DIY motor dynomometer. I've yet to finalize the mechanical design but I am using an ADS1115 in order to read an ACS758 current sensor and a voltage divider so I can record and graph the current over time as fast as possible while the motor spins. RPM is read with a cheap light gate style encoder and a 3D printed encoder wheel, so I can control the CPM. Some motors might be geared down to a few hundred RPM, some brushless motors in excess of 30,000RPM. I'm reading currents in excess of 75 amps at peak which is why I'm using a hall sensor and I also want it to work for much smaller motors too, so I wanted a resolution of ~0.05 amps, hence using the external 16 bit ADC. One problem I ran into is sensor drift. I noticed by just sitting there doing nothing over the course of a couple minutes the ADC read values can drive by up to 50. Other than calibrating to 0 at the start ever each test (which is my current workaround) is there any way to try and stabilize this?
These acs hall sensors are not very good at small currents. Maybe you watch my video concerning current measurements.
Grutti Andreas! In your example at 3:57 , you do not strictly need to apply a clock to the comparators, because only one comparator-value shall change at any given time. What can misfire is the digital truth-table afterwards, (as multiple bits shall change, e.g. going from 1 (binary 01) to 2 (binary10)) - A Gray-code encoding mitigates this too. If the clock can be avoided, the flash-ADC will be even faster.
Thank you for clarification. As you might have seen: I am not a specialist in this matter, just a user...
The Arduino boards *do* provide a pin for the reference Voltage, it's clearly labeled as AREF. The libraries make it easy to use and many beginner tutorials mention its existence.
True. But it is not connected as I said
Hello Andreas! I found your video when searching about adc because i am trying to trick a STM8S003F3 without success.
The STM8S seems to be running at 3,3v and has a 10k linear pot connected to 3,3v VCAP, GND and the variable output to PD3 , which is described as AIN4 and ADC_ETR. It stands for External trigger. From what i understand, the ADC external triggering means it is possible that the ADC only reads on some programmed event. Turns out that when i rotate the POT, even when reading with a digital meter, the voltage values are not linear at all and behaves differently along the POT range. It can read steadily on the starting range, mid range it will fluctuate values and even read 0 and on the end range read some steady values again, higher than on the start range. But the device works correctly. What can be the cause for this? The STM8 adc pin can be programmed to somehow behave like this? Or expecting some pwm signal synchronized with its clock? I don’t have a scope to analyse this in a better way. If i input the POT output to a wemos d1 8266 it will read accordingly to the digital meter, but i cannot put the wemos bypassing that signal to a pwm output back to the STM8, it will not work the same way as if the POT is directly connected to the STM8. Do you have a clue about why this happens?
Thank you
I have no idea. And I never used an STM8. So I am not a big help :-(
@@AndreasSpiess ok. Thanks anyways. Maybe it would be something good to figure out. I gave up on that approach because i don’t own a scope but i really would like to know what’s going on behind the scenes. Cheers!
Very nice characterization fod ADC's.
Thank you!
Thanks for another fantastic video. Just for curiosity, if I insert a delay(1) between each analogRead(A0), the measurements related to WeMos D1 Mini have more fluctuations than without it. Is this due to the capacitor in the Sample and Hold circuit that discharges faster by inserting delay(1)? It is evident within low serial baud rate. Thanks
I too had major issues with the ESP32 ADC. Drove me insane. I'd been using the UNO prior to that with 1.1v reference and was very disappointed with the ESP32 ADC performance. Bullshit indeed lol! Changed to an ADS1115 and very happy. So happy I bought lots of them and won't use the built in ADC now. If you're going to use an ADC, might as well use an accurate one.
Thank you for sharing your expreience!
Seriously good video. Thanks man! Appreciate your hard work :)
Glad you liked it!
Thanks for the video, I had a problem with the adc fluctuating reading with my arduino, I never knew that there are imperfections in these microchips.
I wish you also compared an STM32 board like the black pill (STM32f411)
I thought I did this in another video.
Fantastic as always, Professor!
Thank you!
Here you go, here is the code to help get the ESP32 ADC to behave better:
// configure ADC for getting battery voltage
pinMode(VoltageSense, INPUT); // Battery Voltage Sense input
analogSetPinAttenuation(VoltageSense, ADC_0db); // set the attenuator for the ADC to allow 1.1V input range
analogSetWidth(11); // set the resolution width to be 2048 instead of the default 4096 to try and reduce noise
analogReadResolution(11); // set the resolution to be 2048 instead of the default 4096 to try and reduce noise
Also another trick I found is to add a delay directly before the read, it helps get a more stable reading too! 50ms worked well, but you can’t always do that obviously.
I never tried to reduce the resolution and compare it woth averaging. Good to know. Thanks!
Great timing! I just got ten sliding potentiometers and was wondering how to measure their values. Found another video that shows how by connecting all the value pins with a diode to a single analog input, the VCC pin on the potentiometers can be muxed using digital outputs one after another. Also thinking of using a GPIO extender chip like the PCF8575 for this, so I could measure about 16 analog values (slowly) using a single analog pin in the Arduino with pins left over. Will be a MIDI controller project, don’t need resolution or speed or accuracy all that much imho.
You can buy analog MUX ICs like the 74HC4067
Some of the best educational content out there!
Thank you!
Another problem with multi-channel ADCs on most microcontrollers is that the samples are not taken simultaneously, but interleaved. Better ADC and MUX combinations are possible, that use sample and hold before the mux, so even if the ADC part is doing interleaving, the values being produced correspond to the same moment in time for all channels.
Thank you for clarification
Thank you for this great episode.
Loved it.
Glad you enjoyed it!
@@AndreasSpiess hope you will make one about amp reading too.
How to use shunt to read up to 50amp draw.
Great video. Perhaps in a future video you can explain sigma-delta (or delta-sigma sometimes) ADCs. I think they are pretty cool.
I am not sure if this would be interesting for many people. Maybe more for a specialized channel...
@@AndreasSpiess But two of the ADCs you linked to, ADS1115 and AD7705, are sigma-delta. They're quite common, especially if you're looking for cheap breakout boards.
Me at 16 mins: "ah... bugger" :D Had some "extensibility" ideas in mind for a project to potentially later hook some ADC2 pins up to some inputs... Still, good to know an I2C board is an option if I later want to do go that route and still use Wifi. Video wins just for that, but it had already won several times over for teaching me about how ADCs work!
I am wondering why the chip designer for ESP32 didn’t add shield to the ADC part so that the result will not be influenced by the EM coupling.
Over sampling with noise shaping can increase the precision as well as remove the glitches, which was what I was expecting this video to include. But this method is really a simple and convenient way for the purpose, which is what I haven’t thought of.
I do not know the chip designers. But having Wi-Fi on the same chip is for sure a challenge.
Remember that in Arduino Uno and all the Arduinos with ATMEGA328 you can select an internal stabilized voltage reference, to avoid the errors with power supply fluctuations
This is explained in video #10 ;-)
VRef is applied to the inverting inputs ( - ) of the operational amplifiers. When an input signal (voltage) is applied to Vin (non Inverting Inputs) ( + ), then the operational amplifiers will turn on accordingly because of the rules of the operational amplifier logic as a comparator. The way you have it, if Vin is zero, then all outputs on the operational amplifier is logic HIGH. So LEDs on the outputs on the Operational Amplifiers would turn off instead of turning on when an input signal is applied to Vin.
Impecable, como siempre Andreas
:-)
Nice comprehensive review, one question comes to my mind, is the ESP32-S2 any better than the ESP32?
Notf sure overall, but the esp32s2 has a bit width of 13, compared to the esp32's 12 - which is significant.
I don't think it's shown in the documentation yet? However idf alerts you to this when compiling and it can be seen in the respective header files (of course it works too!).
The same BS :)
@@Sovvyy It's just a marketing. I've tried this to monitor charge of the battery -- it's useless in the same way as regular ESP32 or ESP8266 that ESP32s2 should replace. Use only external battery controllers chips, internal ESP ADC is useless for such sensitive things. And what I hate the most with ESP -- this "filtering" formulas with magic numbers -- you need to adjust it for each and every chip. I never had any 2 chips in ESP32 or ESP8266 lines with the same ADC errors.
Nice video, I always interest in electrical measurement sensor.
However I've noticed you use 2^n-1 for your calculation (eg. resolution, voltage conversion), it's always caught my attention because there 2 side for this all over the internet, whether use 2^n or 2^n-1. To this day I'm used 2^n for my calculation, LSB (resolution) is given by Vref/2^n and voltage sensing using value*Vref/2^n, it's because my understanding that there only 2^n step (value begin at zero) so the maximum value of sensed voltage (FS) is always 2^n-1 sized, not equal or exceed the Vref itself. If the calculation related to FS then I will use 2^n-1 so LSB=FS/2^n-1 because FS=Vref-1LSB.
There so many tutorial, for example 10bit built-in Arduino ADC, using 1023 or 1024. I think it's interesting topic to discuss this fundamental math for ADC.
I had such discussions in my last video where I used 1024. seems to be a complicated topic...
FYI.
Re: ADS1115
It seems to be OK to supply input voltage to the ADS1115 even when it is not supplied with Vcc.
i.e I have a project where I am measuring the voltage of a car battery, using a ADS1115 and a ESP32 , using a voltage divider to reduce max of 16V to 3.3V on the input of the ADS1115
If I remove the supply voltage, the ADS1115 seems to survive even though the input was 3.2V and Vcc was 0V.
Possibly this is because I'm using 4.7k in series as part of the voltage divider, so the current would be limited to about 3.5mA.
In hindsight, I will try using higher values for the voltage divider and use a reservoir capacitor to reduce the effects of the higher impedance input.
e.g. try 47k and 12k as the divider, so that the max current would be only 0.3mA
PS. I just accidently destroyed on channel of ADS1115 by shorting the 4.7k resistor with a solder bridge, hence supplying 16 to one input, but the other inputs still work OK.
So when I rebuild my prototype I will change the voltage divider resistances.
So you were lucky that only one input was destroyed... I try to stay inside the limits of the data sheet.
Maybe the chip has an input protection diode towards VCC and as you write, this protects the chip during power off.
And you are right, you can use higher resistors for the voltage divider.
@@AndreasSpiess
I actually destroyed 2 inputs, because I shorted 2 resistors. However the other 2 inputs still work OK.
Before I replace the module I will do some more testing on the 2 remaining operational inputs to see how much current is taken by the ADC inputs when the battery I am measuring is still connected, but the supply to the ADS1115 is removed, i.e I don't power my entire sensing board.
Like you say, there is probably a protection diode to Vcc on the ADC inputs, but this diode is probably only designed to protect when the supply is removed from the ADS1115 when there is a series resistor connected to the ADC input.
I wonder if someone had a 1.5V battery connected directly to the input of the ADC, whether the input would be destroyed if the ADC was not powered.
I also need to experiment with higher value resistors in the voltage divider, to increase protection, The manufacturer seems to recommend resistor dividers with values around 1k Ohm in total, to prevent measurement error, according to a post on their forum. However the input impedance of the ADC is 3M Ohm, so higher values should not cause significant measurement error and adding a reservoir capacitor will also compensate for the higher resistor values albeit slowing response time.
Bottom line is that the accuracy actually required for a given application drives the decision of if built-in MC's ADC or external ADC will be required. The ADS1115 is my go to solution if it's required, very flexible and cheap.
You are right. This was the goald of the video: To show people the different possibilities with pros and cons...
This was super helpful and timely for me. I've been struggling with how inaccurate the ADC is in the ESP32 trying to get good voltage readings. I'm assuming that I don't need a voltage divider if use the ADS1115 to measure an 18650 Lion and if I power it from the ESP32 it wont cause a current draw when the ESP32 is in sleep mode?
I have learned some useful stuff struggling through this process though, like how to switch on and off a voltage divider using a N and P Channel MOSFET Module (CJMCU-4599 Si4599) so that the divider doesn't constantly drain the battery (thanks Ralph S Bacon). I also looked into using the INA219 board, but the .1 ohm resister creates a constant draw a battery as well. I'm going to give the ADS1115 board a try, I just ordered a few from Amazon. Andreas, thank you so much for presenting this information in such a clear and concise manner. Cheers!
Pay attention with the ADS1115: It cannot measure much above its VCC. So if you connect it to 3.3 you cannot measure 4.2 volts without voltage divider.
@@AndreasSpiess Thanks!
If you made the resistor ladder uneven, and stuck with 4 comparators, you could get useful levels e.g. 3.6 3.8 4.0 4.2v dedicated for monitoring lithium cells
Good idea. But teh ADS1115 is also quite cheap...
The ADS1015 is a drop-in replacement for the ADS1115 and can provide much better stability and speed, but is only 12 bits. Conversion is 3 to 8 times faster when reading values over I2C and sample rate is more than 4 times faster. It's also half the cost and has the same footprint.
Good to know. Thanks!
It's also possible to use a smoothing capacitor on analog values instead of doing too much work in software. I wanted to use a 100K ohm pot going from ground to Vcc and there's not enough current to overcome the internal leakage and you get bad results. Corrected with a 0.1uF capacitor on the pot slider and it removes the error on those internal ADs and you still get the advantage of low current consumption through the 100K pot.
You are right, caps help for outside noise. Unfortunately not for the noise created by the ADCs :-(
@@AndreasSpiess true. I only use the internal A2Ds for basic applications. I just begrudge wasting a lot of current through a pot connected across the power rails. The capacitor method works really well.
It will be nice if there is updated version of this video. Considering now espresif have many varians (s3, C3, C2, etc)
Maybe an idea...
Muchas gracias por tus videos!, son una excelente referencia.
De nada!
At 22:17 it should be: 65535.0 instead of 65536.0
Great video!!!
Look at the comments where they tell me to use 1024 instead of 1023. I do not care because it does not matter if you calibrate your sketch.
@@AndreasSpiess thank you!!!! You are making a great Job!!! Keep on!!!
Really insightful video thanks for posting!
Glad it was helpful!
Nice video :) 2 mistakes.Forget sigma delta adc. And arduino impedence is 10k becose of hold capacitor so voltage divider resistors values is to high
Thanks for the corrrections. I forgot more than just the sigma delta converter.
Fabulous study! Thank you!
Glad it was helpful!
Pretty useful, Mr.Spiess
Glad you think so!