Conversion between mv/g and LSB/g

[oswahhad]

Hello,

I am working on a project as a student in which i need to select a MEMS accelerometer and I am facing a problem comparing between digital and analog sensors wrt to sensitivity. Can someone please help me convert from mv/g and LSB/g ?

Thank you.

 

[KlausST]

Hi,

Impossible without a datasheet.
The information should be found in the datasheet.

Klaus

 

[oswahhad]

Hello,
I have the datasheets but I can only find it either in LSB/g or mV/g.
can you suggest a way to compare the accelerometer sensitivity? I would really appreciate it.

here is an example from the datasheets:

 

[KlausST]

Hi,

I don't understand where the problem is.

The "digital" shows 4 different sensors or one sensor with different range setting.
Let's take the "+/- 4g" setting.
* this gives 8g total range
* and 10 bits resolution means 2^10 = 1024 LSB
So you have a resolution of 8g/1024 LSB = about 8milli-g per LSB
Or the other way round: 1024LSB / 8g = 128LSB/g

And for the analog one: 300mV/g says all regarding sensitivity.
But what is missing: the total range

For both types: you need to have noise and drift specifications to get informations about quality.

Klaus

 

[oswahhad]

Hello,
Thank you for your answer.
I figured out how he calculated it but I want to compare the analog and digital sensors sensitivity, for that i wanted to somehow relate mV/g to LSB/g to be able to compare, like for example if I see 300mV/g and 128 LSB/g, what can I say about the sensitivity, which one is more sensitive.

 

[KlausST]

Hi,

I still don´t see the problem.

again: It´s impossible to compare both sensors without the datasheets.

a digital sensor gives it´s sensitivity in LSB/g and an analog sensor gives it´s sensitivity in mV/g.
They have different output signals, thus they need to give the sensitivity in different units.

Compare "digital" with pieces. Like eggs. You can´t buy half of an egg. If you buy 12 eggs you will get 12 eggs. exactly.
Compare "analog" with "liquids". You can buy any amount of liquids. But if buy "100 ml" you never get "exactly" 100ml, there will always be a difference, maybe very tiny..

Maybe you mean "resoltution"?
Maybe "measurement range"?
Maybe SNR? Or dynamic range?

Please explain.

Klaus

 

[FvM]

The sensitivity numbers have little significance as such, you need to relate it to offset, offset drift and noise magnitude.

 

[oswahhad]

I think i cannot explain my question or maybe since I am a new in the sensors topic I am asking the wrong questions.
If we assume I have 2 sensors one digital and one analog, both with a measurment range of ±200g.
I would like to compare these 2 knowing I have the following datasheets: can you guide me and tell me how to compare?
for example, i can easily compare the bandwidth and the weight (since they have same unit) and choose which one fits better my application,
but for sensitivity, they have different units, i am not able to compare this criteria
same for the noise, in one the noise density is given and in the other one only the noise.
Just if it is possible, a little guidance would be appreciated (by the way, this is just an example, it is not my project work, i am looking for a sensor in a different range).

Thank you for your patience and I am sorry if my question seems unintelligent.

and this for digital:

 

[KlausST]

Hi,

Just if it is possible, a little guidance would be appreciated

I told you twice now that it´s impossible without datasheets. But you still hide them.
Instead you show two pages of different datasheets. Not the sensors before.
Both of digital sensors.

I also tried to tell you the difference between analog and digital and why they need different sensitivity units.
Just to clarify: You can´t directly compare random digital and analog sensitivity values without additional information.

On digital sensors there is no "millivolt" you can refer to, and on analog sensors you don´t have LSBs.
You may compare "g", since both have them.

Or you may use an analog sensor and an ADC to get LSBs...
Or you may use a digital sensor and an DAC to get millivolts...
But in both cases you need the ADC/DAC transfer function to know how millivolt relate to LSBs

Klaus

 

[oswahhad]

Hello,

I really understood it when you said it is not possible to compare directly and that I need more information.

That is why what I am asking is an alternative on how to do so (on what to focus to compare analog and digital sensors with just referring to the information in the datasheets) and I am very sorry I made a mistake uploading the photos.

 

[KlausST]

Hi,

Are these two new devices?
So we now have (parts of) datasheets of at least 5 different devices. .. to increase confusion.
*******
Let´s focus on the parts of post#10.
At least two accelerometers with identical measurment range and both are for 3 axis.

ADXL377
* has analog output.
* has an ouput offset of typ. 1.5V
* has a sensitivity of typ. 6.5mV/g
Example: +50g.
50g x 6.5mV/g = 325mV
add offset: 1.500V + 0.325V = 1.825V (This is the expected output voltage for +50g)

ADXL375
* has digital output
* has a sensitivity of typ. 20.5 LSB/g
Example: +50g:
50g x 20.5LSB/g = 1025 LSB (this is the expectable output of the digital interface for +50g)

*******
Please answer these questions:
1) Do you know what "sensitivity" means?
2) Do you understand that "sensitivity" is no measure of quality?
3) Please describe the goal of your application:
[acceleration] --> [dont care for now] --> [goal]
is your goal: Measurement / regulation / processing (analog or digital) / archivate / ???
4) what are your requirements regarding accuracy/precision/resolution/ noise/ bandwidth

Klaus

 

[oswahhad]

Hello,

Thank you very much for your time and the clarifications.
Everything is managed now.

I have decided on working with an analog ADXL sensor and an MCC118/128 as a measurement module.
Since I will use the sensor for condition monitoring, my next step is to find methods to compensate/enhance the low MEMS performance (compared to costly sensors).
I thought of maybe using 2 or 3 MEMS accelerometers at the same time. do you think it is possible to get better result? Or maybe you can suggest other possibilities/methods.

I read a lot of research papers and they are mostly talking about sensor calibration but I am looking for a more online method (no calibration..).


Best regards,
Oswa.

 

[Aussie Susan]

You probably need to start from 'the other end': what resolution/specificity/accuracy/whatever-you-want-to-call-it do you need?
Also what are you going to do with the signals you get from the sensor? Will you be processing them in a fully analog manner (in which case the digital one will need to be passed through a DAC) or a fully digital manner (in which case the analog one will need to be passed through an ADC) or what?
Normally design starts with the requirements and the high level design approach, with the part selection coming much later. You seem to be starting with the part and then looking for a way to use it!
Susan

 

[oswahhad]

It is not that I am starting the other way around but this is what the project requires. the project consists of using MEMS accelerometers, a measurement device and a Raspberry Pi in order to find a different/low cost alternative to monitor systems and yet deliver good performance. So the objective is to find a low cost solution for condition monitoring (it is a research project).

Everything will be processed in an analog way; the low cost MEMS sensor is analog, the measurement module receives analog inputs and then converted to digital using ADC.
and my task now is to find a method to compensate/enhance the performance.

As I mentioned before, I am a student, I am really interested in the field and I am trying to learn.

 

[d123]

Hi,

You (unfortunately) probably gain a better perspective on A or D type by doing an error budget for each part (summing all errors and turning the total to a percentage, it looks quite hard/laborious), I think that was already mentioned.

Do your datasheets show minimum, typical, and maximum for the sensitivity of the analog and digital versions? ...As this seems to be the main feature you're interested in for now... Perhaps from those figures, if available, you can work them out as percentages and compare that way. You know: 19 min to 21 max lsb with 20 as typical (ideal) per g is 19/20 = 0.95 or 21/20 = 1.05, i.e. - or + 5% either way; then 5.8mV/g to 7.2mV/g for typical 6.5mV/g would be 5.8/6.5 = 0.89, 89%, -11% and the other 1.10, 110%, +10%.

Not sure if valid method for comparing these two ICs after reading sensible comments about eggs and liquids, etc., but it's a suggestion, because presumably 20lsb/g and 6.5mV/g amount to the same thing.

Maybe you can think of comparing inches to centimetres and which has better resolution for measuring 5mm or 50mm - is that what you mean about sensitivity?

Or (assuming 20lsb per g) something like instead of how many grams, how many milligrams does 1lsb cover compared to how many milligrams per 6.5mV/20? It's the same as mV or lsb per gram, but more detailed, maybe?

These are just suggestions that may be useless or wrong for some reason or another, just suggestions of possible comparison methods of apples to pears.

 

[KlausST]

Everything will be processed in an analog way; the low cost MEMS sensor is analog, the measurement module receives analog inputs and then converted to digital using ADC.

This sounds contradictory.
If it is processed in analog way ... then you don´t need an ADC.
You need an ADC if you wan to process it in digital way.

I still have no clue about what really is the task,

Klaus

 

[BradtheRad]

Sensitivity can say something about the smallest motion, or acceleration, that yields a usable reading from each type of sensor. Or unambiguous reading. Or a stable reading.

I picture doing real-life tests on the sensors simultaneously side-by-side. Do slow movements, jittery movements, angular tilting, etc.

 

[danadakk]

Maybe some confusion on my part. One "normally" is told what the goal is,
and then starts looking at sensor costs/performance to achieve that. Then
one starts looking at the signal conditioning path and ultimately the
processor/AtoD to move / process / interface the sensors to complete the
goal. These choices can be interactive, one design area influencing another.
But all driven by basic system goals.

You seem to have the sensors already picked, same with processor.

So one now looks at Vout of sensor and range / resolution / accuracy / speed of
A/D needed to convert the sensor signal into digital. This decision also can be
interactive, one influencing the other.

If you attempt to do this by designing a signal path with no goals one can
simply use a wire and thats all that is needed. After all thats the cheapest
solution not knowing what the signal is, not knowing the other design goals
in accuracy, repeatability, resolution, environmentals, packaging.....are.

If the task is given without goals then you need to find a new client, or a new
school if thats the best discipline they can offer in their education methods.


Regards, Dana.

 

[d123]

Hello,

Thank you very much for your time and the clarifications.
Everything is managed now.

I have decided on working with an analog ADXL sensor and an MCC118/128 as a measurement module.
Since I will use the sensor for condition monitoring, my next step is to find methods to compensate/enhance the low MEMS performance (compared to costly sensors).
I thought of maybe using 2 or 3 MEMS accelerometers at the same time. do you think it is possible to get better result? Or maybe you can suggest other possibilities/methods.

I read a lot of research papers and they are mostly talking about sensor calibration but I am looking for a more online method (no calibration..).


Best regards,
Oswa.

Hi,

You'll be lucky to get away with that method . Think about it a bit - part of the information from the sensor(s) will be junk (inaccurate, 'wrong'), might make next stage do - or not do - something when it should(n't). Maybe your requirements are loose and accuracy/precision is not so important.

Op amps and a quality (low ppm/C) trimpot come to mind rather than averaging three sensors' outputs (through a summing amplifier).

Is sensor error linear with temperature, convex or concave? If it's 's' shaped, I wish you well and wouldn't have any ideas on curvature-correction (without an MCU and code to handle it).

A cheap analog temperature sensor output passed through a resistive divider to get required mV/C (for a linear error) + OA as difference amplifier to subtract error from sensor output could do linear error. Maybe, only maybe, temperature sensor + diode + diff amp (+ inverting amp depending on error curve) might compensate for parabolic error but would be kind of exponential in nature so not a great fit.

I think people use MCUs for 'no human calibration required' in mass production. For only one or a few boards, presumably/possibly a trimpot might help.

I realized last night as well, a) at least one of my comparison methods is useless, and, b) probably what the sensor signal is fed into matters more than sensor sensitivity - an ADC that can only sense with 'accuracy' or 'granularity' or whatever the word is with +-1mV error is already going to be possibly ~15% wrong for each 6.5mV/g, or an OA or comparator with 3mV to 10mV input offset will not see gram #1 (and part of gram #2). What is the sensor fed into?