Would youplease explain the difference between "noise density", "output noise" and "Velocity Random walk" parameters in your spec sheets (i.e. adis16407)

My best understanding is as follows. I will greatly appreciate if you can comment on it:

1. Noise Density: this is the (flat) noise spectral level of the analog sensor output which is later discretized in the front end.

2. Output noise: This is the rms noise of the sampled signal (probably after aliasing filters but before any user specified output filter stage).

On the other hand, the numbers that are indicated in the spec sheet do not comply with these definitions. Figure 19 of ADIS16407 spec sheet says the internal BW of accelerometer is 330Hz. In this case the rms output noise should be:

0.5mg/sqrt(Hz)*sqrt(330*1.6)=11.48

However the speec sheet says is is 9mg. (page 24)

3. What do you mean by "velocity random walk". Apparently, you use the value of Allan variance figures at tau=1 for VRW. (This is also how it is defined in the usual texts). However, in your case this is completely wrong. You have a huge quantization step, which is 3.33mg. Basically your stationary output is nothing but the quantized white noise. The Allan variance of the quantized white noise is completely different (I would rather say arbitrary) than the allan variance of the unquantized noise. Therefore, your apperant use of the term "velocity random walk" is wrong. As a matter of fact, because of the very large quantization step, I doubt that your Accelerometer Allan variance figure (i.e. Fig.8) represents anything meaningful.

(PS: Quantization error defined for high-end inertial sensor is a completely different phenomenon which manifests itself as a -2 segment in allan curves. I mean "bit quantization" in the above explanations as your sensors do not have an "Qunatization error" in the sense that we use in high-end sensors.)

Hello ode45. Thank you for the clarification. I appreciate your insight on this and believe that you are correct. I plan to offer a more generic post for this, but wanted to provide a clear statement that all characterization on noise/stability/random walk is performed using SMPL_PRD = 0x0001 and assumes the following:

I am going to offer this clarification in future revisions of product documentation, which are impacted by this feedback, which include the ADIS16334, ADIS16385, ADIS16407 and also in future products that employ this architecture. It is worth noting, that the ADIS16133, ADIS16135, ADIS16136, ADIS16375 and ADIS16488 all provide this same function but also provide access to the extra bits (through additional register), so we would expect that they would provide optimal performance. Thank you again, for offering this feedback and for your interest in these products.