Load cell with HX711 and datasheets

The datasheet of HX711 is found here:
https://www.digikey.com/htmldatasheets/production/1836471/0/0/1/hx711.html

Furthermore, I have looked at some of the cheaper load cells with an aluminum bar as you see here, and I found similar data on all of them:
https://cdn.sparkfun.com/datasheets/Sensors/ForceFlex/TAL220M4M5Update.pdf

The HX711 integrated circuit is mostly sold mounted on a small PCB together with cells. When load cells are described in this forum the HX711 is used at least half the times. I have no practical experience with them, but when I look at the data sheet, I am not that impressed. I think that it normally be only a small part of the AD-converters dynamic range that is utilized, and therefore I expect more noise on the measured values is introduced compared to better alternatives. Am I right?

This is a theoretical approach from reading – I know.

The AD-converter in the HX711 does have a huge dynamic range of 24 bits, but I guess that a significant amount of noise will be there on the less significant bits including some other errors. 24 bits AD-conversion are normally very difficult to achieve, and this is a low-cost integrated circuit. Therefore, I guess that the results are far from that accurate.

Assume that you have got an excitation voltage to the wedstone bridge of 3 V, and the same voltage is used for AVDD of HX711. Then the expected rated full scale output range is +/- 3 mV for both positive and negative forces on the load cell. The internal amplifier can be set to a maximum amplification of 128, and therefore the input to the AD converter is +/- 384 mV. Consequently, you use only 100 * 2 * 0.384/3 = 25,6 % of the AD converters dynamic range. When you use the load cell with only zero and positive loads, then you only utilize half of that eg. 12,8% of its range.

The load cell will typically be damaged if you apply above 150% of the rated load for its full-scale output. In most applications you can expect significantly higher short-term dynamic loads due to blows or some perhaps unexpected things happening.

As an example, I got an application to measure the pressure applied to a pedal to be used for speed control of a sewing machine. I think you might expect a down force from a foot up to 30 kg (300 N) in some few lifetime cases. But in normal use the max speed would be, when you press 4 kg or 40 N. Consequently, I like to choose a load cell specified to 20 kg. But then only 10 % of the full-scale range of the load cell will be utilized in this application. And it will then be only about 2.5 % of this AD-converters range.

With the HX711, you can choose to measure the input every 12.5 ms or every 100 ms. The datasheet does indicate that you will see more noise when you choose 12.5 ms. I know that you will need a fast response for the pedal, and 100 ms will not be acceptable. In this application you cannot apply digital filters to reduce noise.

As an alternative for this application, I should design a circuit with an op-amp with an amplification of about 2500 with manual zero-point-adjust. The output can then be measured by a normal analog input of a microcontroller.

Have you been able to achieve fast and accurate measurements with load cells using the HX711? How accurate is this AD-converter? (I do not find data on accuracy in the datasheet).

I made a coffee scale and it is fine. It is instantaneous reading as far as I can tell, and I limit it to 1 decimal as my display is fully utilized. I see lot's of folks using it. You sound like you know a great deal about it, so I am surprised you are asking for help.
My latest upgrade is I also now disoplay the number of beans needed to get to my zero point.

The voltage regulator on HX711 boards is set to 4.3volt.
I have seen practical resolution limits of the HX711 of 16 bits mentioned.
Make sure you use a 5volt-logic Arduino, or a Sparkfun board if you use a 3.3volt MCU.
Leo..

I know a commercial application that uses a bubble-type foot switch and a pressure sensor.

The designers used that approach because the option with the load cell was too complicated to implement. It had the problems you mentioned and many others, found in the development process.

Thanks. When the load cell excitation voltage and the internal voltage, AVDD for the AD-converter is the same (it is in all the schematics I have seen) then you get the same amount of the AD-converter range covered. Anyway I have seen different schematics regarding this voltage, and I think most of them provide 4.3 V with 20k and 8.2k resistors, as you mention. However I have seen others too like these:
https://electronics.stackexchange.com/questions/696330/hx711-analog-supply-regulator-current-draw-in-sleep-mode
(provides 2.9 V)
https://easyeda.com/modules/HX711-Breakout-Schematic_cbe4869f910549ab82bd3c6fe35b77e8
(provides 2.5 V)
Therefore I was not quite sure what was normal.

16 bit sounds reasonable and 256 times lower than 24 bit. But 16 bit is still very accurate. I wonder if the design perhaps use a bit other block diagram, so they incorporate an additional programable amplifier to extend the dynamic range at low signal levels.

Yes, Elna and Singer had some sewing machines made in the 1980ies with air pressure pedals. Look here for the Elna bubble type:
https://www.ebay.co.uk/itm/145692469172
Many users are still fond of them.
I have used this principle for a DIY design as well, and it do work well.
But the principle do have a problem, because air expands by temperature due to the ideal gas law, and you got a confined volume of air being used. So the zero-point change significantly with temperature. It is possible to cope with this problem with an accessible manual zero point adjustment or by some software recognizing the zero point changes, because you do not sew most of the time.

E+ and AVDD must be the same, to preserve ratiometric behaviour between load cell and A/D. VCC of the digital part can be different.

Maybe excitation was dropped to 2.9volt to make it compatable with a 3.3volt supply and MCU.
Leo..

1 Like

True. So that should be a clue to it's accuracy or lack of.
I would guess that the noise free resolution is around 16 bits.
How much more resolution you loose due to nonlinearities is unknown

This topic was automatically closed 180 days after the last reply. New replies are no longer allowed.