Make a slight modification to the HX711 PCB for better measurements #117
Replies: 8 comments 51 replies
-
|
Can confirm, less noisy after mod! Great find!! |
Beta Was this translation helpful? Give feedback.
-
|
Excellent! Glad to hear! |
Beta Was this translation helpful? Give feedback.
-
|
This is an interesting mod! I've been using my Chinese HX711 successfully for the last 4 months, and have had great success. I'm curious if this modification would improve accuracy and repeatability with lower noise from the sensor. Im not sure its worth the effort in my case, but would love to see before and after noise readings. |
Beta Was this translation helpful? Give feedback.
-
|
Well, it will certainly improve performance, but whether that translates to a tangible benefit depends on a lot of variables. If you're not having issues as-is, you likely have a relatively "clean" 3.3V rail driving the VSUP pin on the HX711. If you're happy with the repeatability and precision, I wouldn't bother making the change. |
Beta Was this translation helpful? Give feedback.
-
|
This completely fixed all my problems with my Specialita and the 1kg Mavin load cell! Thanks for this valuable information. Should be at least mentioned in the guide IMO. Note: I didn't have a 12k resistor, only a 10k one. Voltage drops to around 2.9V with that, but still it completely solved my noise issues. I was barely able to calibrate. Now on first try. Also before I got the warning indicator on the display showing detected noise. Since 2 days I have this mod I didn't see it yet. Awesome. |
Beta Was this translation helpful? Give feedback.
-
|
You're very welcome! 2.9V will be totally fine, a reasonable 400mV of headroom when using a 3.3V VSUP, and still well above the minimum voltage (2.6V) permitted on AVDD. Oddly, the configuration of the AVDD linear regulator on HX711 implementations is not well documented anywhere in the product pages of HX711 breakouts, seems a lot of people missed this detail when designing and using the HX711 IC. |
Beta Was this translation helpful? Give feedback.
-
|
The downside of this solution is that you lower the excitation voltage even further (the loadcells are rated for 9-12V excitation), which lowers the signal to noise ratio. Another approach is to supply the HX711 with 5V and add a 1K:2K voltage divider to the data pin. That way you still have a clean 4.3V excitation from the HX711 and around 3-3.3V data output for the ESP32. Another option is to use the more expensive sparkfun board with separate 5v and 3.3v supply. FWIW I used the red shielded pcb version. I don't have the green one; so couldn't test if it results in any differences. But at the test setup it gave good consistent results (even with quite a long cable to the loadcell). |
Beta Was this translation helpful? Give feedback.
-
|
I disagree with your assertion that it lowers the signal to noise ratio. Broadly speaking, this modification raises the signal to noise ratio, which is why users (myself included) report better results after completing this mod. By eliminating the prevailing system noise caused by operating the HX711 linear regulator incorrectly, the primary source of noise is reduced, resulting in an overall better SNR. Additionally, given that the full-scale differential input voltage of the HX711 is a function of the AVDD voltage, users are not "missing out" on available ADC counts, because the ADC reference scales with the excitation voltage. I do agree that excitation voltage should not be reduced without good reason, as doing so does bring the system closer to the noise floor of the entire system, and this does indeed typically result in lower SNR. It also reduces the size of of each least significant bit at the ADC, which makes the system more susceptible to other board level noise sources. However, in this case, the pros of fixing the HX711 linear reg outweigh the cons of lowering the excitation voltage. If you want the best system with the HX711 breakouts widely available, you should use the Sparkfun board, and run a separate 5V net down to the board so that the default 4.3V AVDD can be used, but I find this unnecessary for the measurements that we're making for coffee bean weighing. |
Beta Was this translation helpful? Give feedback.
-
|
Ok, maybe rephrase my statement about the SNR. I was talking about the difference with the 5V operation (as stated in the specs of those boards) instead of 3.3V. If you can only feed it with 3.3V then the modification certainly provides a better result. However, the sensitivity of the loadcell is 2mV/V. With 2.9V applied the signal received is 0-5.8mV whereas 4.3v results in 0-8.6mV. The latter will result in a better SNR. The problem with the cheap chinese breakouts is that the datapin will have the same voltage as the VCC, which is a problem for the ESP32. So you could add 2 resistors to solve that problem (1k between data and GPIO and 2k between gpio and ground), without sacrificing SNR. Both solutions will result in a system which works just fine, I didn't measure or test the differences. So I chose the 1k:2k divider route purely based on theory. |
Beta Was this translation helpful? Give feedback.
-
|
That makes sense! Yes, you can choose to supply 5V to both DVDD and VSUP to use the default 4.3V AVDD setting (and these two supply pins are often tied together on cheap breakouts), but as you've correctly called out, you'll have to level shift the DOUT signal downwards to be compatible with the 3.3V logic used by the ESP32. Seems like people get away with leaving the clock signal as-is at 3.3V logic levels, but it's not clear exactly what the VIH-MIN threshold is on the HX711 as the datasheet does not specify. If it's the standard 0.7*VDD for typical CMOS processes (which with a 5V VDD, is 3.5V), then I'd worry about the 3.3V clock signal having enough amplitude to reliably clock in as a logic HIGH. If I were going to go with this approach, I'd probably slap a proper logic translator in the system (such as https://www.adafruit.com/product/1875), and translate both the clock and data lines. |
Beta Was this translation helpful? Give feedback.
-
|
As I'm not too deep into the schematics I tried to identify R1... There isn't one at my HX711 pcbs: 
To add to the confusion: I plugged an oscilloscope to GND and VCC - what a mess. Very frequent peaks and drops, way from stable. Would a capacitor help ? |
Beta Was this translation helpful? Give feedback.
-
|
Thank you @DC01. I can't solder SMD components that small, and I don't have the SparkFun PCB. Where can I get one? Currently, I’ve connected a 5V HX711 (from AliExpress) and used a step-down converter to 3.3V for the DOUT pin. This actually fixed the negative weight problem! However, after about 5 hours of runtime, I start getting a yellow rectangle error indicating a noise issue. A reboot brings it back to normal temporarily. Any ideas? |
Beta Was this translation helpful? Give feedback.
-
|
Where are you located ? |
Beta Was this translation helpful? Give feedback.
-
|
middle east :( |
Beta Was this translation helpful? Give feedback.
-
|
Ah.. bombastic environment ;-) AliExpress: Soldering iron suitable for SMD (buy the one with case) AliExpress: SparkFun Load Cell Amplifier-Hx711 |
Beta Was this translation helpful? Give feedback.
-
|
I will try to look for it locally; otherwise, I'll order it from Amazon or AliExpress. Hopefully, I'll be able to stabilize the noise a bit. The T70 load cell wires are just so thin that I'm not sure they soldered well this time. Why can't they use better wires? |
Beta Was this translation helpful? Give feedback.
-
|
That only works if you stepdown the voltage of the data pin. Otherwise you’ll damage the esp. See discussion above.
|
Beta Was this translation helpful? Give feedback.
-
|
Not quite. The ESP32 board with the display needs to be powered off of 5V, as it has its own 3.3V regulator (which needs a higher voltage input source to operate) that runs much of its circuitry. However, you can also use the 3.3V supply from the ESP32 board to power other stuff (like the HX711 breakout). So at the end of the day, you'll have access to 5V, and 3.3V to potentially power the HX711 breakout board. The main issue we've been discussing here is that the discrete linear regulator on the HX711 breakout either needs to be supplied with 5V, or it needs to be adjusted to work correctly with a 3.3V supply. The problem, is that if the HX711 is ran off of 5V, the digital output signal needs to then be down-adjusted to fit the 3.3V logic range of the ESP32. I think the easiest solution is to just fully run the HX711 breakout off of 3.3V (use 3.3V for both VDD and VCC on the Sparkfun HX711 module if you have it, or on a Chinese clone, use 3.3V on the single supply pin), and make the small resistor modification to the HX711 breakout to allow its discrete LDO to work correctly. This is accomplished by simply swapping the 20k resistor for a 12K resistor on the HX711 breakouts. There's some pictures in this discussion that help identify which resistor needs to be swapped. Alternatively, if you have the Sparkfun HX711 breakout, and don't want to do any SMT soldering, then just run 5V to its VCC pin, and 3.3V to its VDD pin, and you're good to go. |
Beta Was this translation helpful? Give feedback.
-
|
Alright, thanks alot for the clearing up, so as i understand it in laymans terms: The HX711 clone is usually ran off of 5v hence the resistor 20k to drop that to 3.3, but when powering via 3.3 we decrease the resistance to ensure we still get the correct voltage for the LDO to work? Heres the schematics of the one i bought, are you able to figure out if its "as good" as the sparkfun, or needs the mod? 
|
Beta Was this translation helpful? Give feedback.
-
|
They don’t specify the resistor values in this diagram (R1 is the target relative to R2). Do you have a picture of your board ? |
Beta Was this translation helpful? Give feedback.
-
Yes and no. The HX711 close is indeed usually ran off of a single 5V supply. The default setpoint of its discrete linear regulator (a set of circuitry on the HX711 breakout designed to provide clean power to the load cell) is 4.3V, not 3.3V. The 20k resistor (R1 in your schematic diagram) in conjunction with the 8.2k (R2) resistor set this discrete linear regulator to 4.3. Due to this default setting of 4.3V, you can't supply the HX711 breakout with a single 3.3V supply, because the supply voltage for this discrete linear regulator must be higher than its setpoint. When the 20k is swapped out for a ~12k, this lowers the setpoint to ~3.1V, which then allows you to run the HX711 breakout from a single 3.3V supply (because 3.3V is greater than the 3.1V setpoint of the discrete regulator on the HX711 breakout). |
Beta Was this translation helpful? Give feedback.
-
|
This one is not "as good" as the Sparkfun board. It does not have the option to run VCC and VDD off of separate supplies. You will need to modify something. I recommend that you identify the 20k ohm resistor, and swap it for a 12k ohm resistor. Then you can run it at 3.3V, and you're off to the races. |
Beta Was this translation helpful? Give feedback.
-
|
This is the best I got: |
Beta Was this translation helpful? Give feedback.
-
|
So I just reflowed all joints. Resistor is 123/12k, I get 2.96v on E+ and 3.27 on VCC |
Beta Was this translation helpful? Give feedback.
-
|
I still get a std dev somewhere in the neighborhood of 0.15g, but it's very change. |
Beta Was this translation helpful? Give feedback.
-
|
Any other ideas? |
Beta Was this translation helpful? Give feedback.
-
|
Ive now ordered a sparkfun HX711, and am hoping that that will resolve my issues. Im guessing I can keep the wiring the same and still power it via the 3.3v from the esp right? |
Beta Was this translation helpful? Give feedback.
-
|
Yes, tie VDD and VCC together (see image below), and power it off of 3.3V from the ESP. Take a look at my wiring below, and make sure yours looks the same. Did you ensure your GND connections are tidy? I tied all of mine together, and then soldered the bundle to the "G" pin next to the 5V pin on the ESP32 module. 
|
Beta Was this translation helpful? Give feedback.
-
|
I did it very similarly, but I will be changing all wiring out for thicker gauge, properly try to give them a few turns to minimize the interference before crimping and in the same go replace the HX711. Maybe that will work. |
Beta Was this translation helpful? Give feedback.
-
|
When twisting the wires use clk + gnd and data + +V to have twisted pairs and then twisting them all together 😉 |
Beta Was this translation helpful? Give feedback.
-
|
Extremely unlikely that you're getting data corruption on the digital signals. Slow speed, high amplitude SPI signals are very resilient. More important than twisting, is to make sure the physical length of signals is minimized (use a short of wires as you can between the HX711 and ESP32). Similarly, thicker gauge will likely provide no help. The HX711 is a very small load to the ESP32 board, less than 2mA, you could power the whole thing with 30 AWG. Make sure the wires between the load cell and HX711 are as short as possible, and routed away from other electronics. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
This is actually something that all implementations of this mod would benefit from. In my implementation, I used the Sparkfun HX711 PCBA that exposes two separate voltage rail inputs. However, I wanted to run a single voltage supply to this board so that I could use a 4 pin connector rather than a 5 pin connector, but this necessitated operating the HX711 off of a single 3.3V rail rather than 3.3V for IO and 5V for the on-board LDO's excitation rail. Without modification, the board will still work, but the excitation LDO will be railed out against the 3.3V supply, and will be much noisier than it needs to be. Luckily, this is easy to achieve, you just have to down-adjust the setpoint of the LDO from it's default value of ~4.3V, down to something at least 200mV below the 3.3V rail. This was easily achieved by swapping out R1 (20k ohm) for a 12k ohm resistor, which then sets the LDO to ~3.1V. Now VCC and VDD can be powered off of a single 3.3V rail.
Many folks who implement designs with the HX711 buy Chinese designed boards that don't even separately breakout the VCC and VDD pins. On these boards, it is even more important to perform this modification, because there is not even the option of providing a separate 5V supply to the HX711 PCBA.
Here's the story on the HX711 as I understand it:
One of the original breakout implementations for the HX711 was made by Sparkfun, and they intended for users to use it in a dual supply configuration. A 5V supply would be used to drive the input to the HX711 controlled linear regulator, and 3.3V or 5.0V would be used to set the interface voltage. As they assumed users would have a 5V rail available, they locked the setpoint of the HX711 LDO to 4.3V via the 20k and 8.2k ohm resistors used. This worked great, until Chinese clones began popping up that omitted the VCC pin (choosing to internally tie it to VDD). They unfortunately did not make any adjustments to the LDO circuit, using the same 4.3V setpoint. Now, if you operate one of these poor HX711 implementations off of a single 3.3V supply, this excitation LDO is still trying to produce 4.3V as per it's feedback configuration, but will never achieve this goal as it is supplied by 3.3V. The effect is that it effectively turns into a pass-through for the 3.3V rail, and you've negated all the benefits that a separate low noise excitation rail had to offer.
Really easy to fix, you just have to down-adjust the setpoint to something suitable for a 3.3V upstream supply level! The HX711 itself supports a 3.1V LDO setpoint no problem (and actually only requires 100mV of dropout from the upstream supply - you could do 3.2V, I just wanted a bit more margin). There was no reason it needed to originally be 4.3V, other than that a larger excitation voltage typically equals more resilience to noise.
TLDR: Swap the 20k resistor for a 12K resistor if using a single 3.3V supply on HX711 boards!
Beta Was this translation helpful? Give feedback.
All reactions