Accelerometers comparison

[dburgt98]

Dear community,

I am selecting an accelerometer to use in a project and I am looking for some advice.

Essentially I am looking for the highest performance accelerometer (regarding noise) in the STM portfolio.

The sensor requirements are: ~±2g scale support, I2C support (preferred), FIFO buffer, at least 1 (absolute) threshold interrupt. 

I am currently using the LSM303AGR in my design, which I am comparing to the LIS2DW12 and LSM6DSV16X as a potential replacement. For the comparison (see table), the most important factors are noise and active current consumption in high-performance mode. Note that the LSM303 and LSM6 are not pure accelerometers, since they offer additional functionalities (i.e., magnetometer & gyro). I am only interested in the performance of the accelerometer part. 

Sensor	Noise	Active current consumption	Costs component (eu.mouser.com)	Costs breakout board (eu.mouser.com)
LSM303AGR	3mg (RMS)	40µA@200Hz	€4.15	€11.88
LIS2DW12	90μg/√Hz	120µA	€1.24	€3.71
LSM6DSV16X	60μg/√Hz	190µA	€3.80	€18.95

To me, the LSM6DSV16X seems to have best performing accelerometer regarding noise.

However, for the LSM303AGR, the noise is only specified as RMS in the datasheet, but not as μg/√Hz.
How do I know the μg/√Hz ratio for the LSM303AGR, so that I can appropriately compare? 

The LSM6DSV16X and LSM303AGR are slightly pricier than the LIS2DW12. I expect this is due to the additional functionalities (gyro and magnetometer respectively), which I do not require in my design.

Except for the costs, could there be any reason to chose the LIS2DW12 over the LSM6DSV16? 

When looking at the STM portfolio for sensors with lower noise, I have also found the LIS3DHH(45-60μg/√Hz), which does not have my preference since I prefer I2C support. There is also the IIS2ICLX(15-30μg/√Hz), but I believe it is designed specifically for inclinometers, it is pricy and has a rather high power consumption (420µA). There is also the AIS25BA(30-50μg/√Hz, but it does not support a FIFO buffer.

Perhaps I am missing out on other low-noise alternatives that do meet my listed requirements, so any suggestions are highly appreciated.

Any feedback is highly appreciated.

Kind regards,

Dirk

[Andrea VITALI · Accepted solution]

Very nice job! Seems you have all the information.

IIS2ICLX is only 2-axis, the lack of the 3rd axis may make it unsuitable for your use case. Also, IIS2ICLX has a ceramic package, which makes it very stable: temperature induced effects are very small compared to other accelerometers in plastic packages. Also, aging effects are greatly reduced because again ceramic is less sensitive to humidity.

RMS noise = noise density * sqrt(bandwidth)

So, one could reverse the formula to compute the noise density given the RMS noise.

In high-performance mode noise density matters because after ADC averaging and down-sampling is performed. And further digital filters can be enabled to further reduce the bandwidth and reject out of band noise. So you can estimate the final RMS given the bandwidth of the filters you have enabled.

LIS2DW is an accelerometer only optimized for low-power consumption and that is why it offers so many low-power modes.,

LSM6DSV is an accelerometer with gyroscope and a ton of features. It is has been optimized for high-value features, not for low-power (even if it does have low-power modes). It has sensor fusion (SFLP), machine learning core (MLC) to process the data extract features and run decision trees, finite state machine (FSM) to detect specific sequences of events, and so on and so forth. It is capable of supporting optical-image stabilization and electronic-image-stabilization with dedicated processing chains. It has a dual accelerometer acquisition chain so you can measure acceleration with high accuracy and low-full scale and at the same time get lower resolution data with 16g full scale with less risk of saturation. It is a beast.

[dburgt98]

Hello Andrea,

Thank you for your clear explanation!

From your explanation, I take that the LSM6DSV  is basically the superior option compared to LIS2DW, though it has higher power consumption. So that is clear. 

You mention that IIS2ICLX is unsuitable for my use case, which I have not mentioned much about. The use case is about detecting short displacements (up to 11cm) in a specific direction, through double integration of acceleration values. If I wanted to detect displacement in only the Y or X axis, would the sensor then be a suitable choice?

As for the LSM303AGR, I am currently not applying the high-pass filter to the output registers. So is there a default bandwidth value that I would have to enter in the formula you provided? Is this related to the sampling frequency?

Kind regards,

Dirk 

[Andrea VITALI]

Correct I do not know about your specific use case, which is why I wrote "may".

Yes, if displacement is only for X and Y, then IIS2ICLX would be ok.

For double integration, aka dead reckoning, you may want to check Design Tip DT0106.

For the bandwidth, at most it is ODR/2 (output data rate divided by 2). For example, if ODR is 100Hz, then bandwidth is at most 50Hz. If some low-pass filter is enabled, bandwidth can be a fraction of that (ODR/10, ODR/20, or lower).

High-pass filter should have a very low cutoff frequency for accurate double integration. In practice one may want to just subtract the DC component (just subtract the offset as measured in steady conditions) and then perform double integration on the result.

[dburgt98]

Hi Andrea,

Thank you! I will definitely check out your suggestions.

I have a final question regarding the calculation.

For 100Hz, I am now assuming the following: 

3mg (RMS in datasheet) = noise density * sqrt(100Hz/2)

3 = noise density * 7.07

Noise density = 3 / 7.07 = 0.42mg 

Am I interpreting you currently? 

Kind regards,

Dirk 

[Andrea VITALI]

Correct noise density as computed from RMS at 100Hz would be 420ug/sqrtHz

Actually, keep in mind that the figure on the datasheet is typical, not the worst case. So 500-600ug/sqrtHz would be more realistic.

And the structure is not fully symmetrical, therefore X may be different than Y axis, and the out of plane axis Z may be worse. Which means that you can should budget for 600-800ug/sqrtHz.

LSM303AGR is a relatively old component, it is an accelerometer and magnetometer combo. The accelerometer in there is few generations behind the LSM6DSV. You can see that the noise density of the DSV part is one order of magnitude lower.

[dburgt98]

---

@Andrea VITALI wrote:

LSM303AGR is a relatively old component, it is an accelerometer and magnetometer combo. The accelerometer in there is few generations behind the LSM6DSV. You can see that the noise density of the DSV part is one order of magnitude lower.

---

This is exactly the reason that I was looking for a newer component. 

Thanks again for the assistance. I don't have further questions.