INA849: Maximum noise specification

Machida-san,

First, that statement needs a limiting factor of a specified gain, as in a gain of 1, the noise is almost 10uVpp RTI, and in a gain of 1000 the noise is roughly 40nVpp. I have included the typical graphs from the datasheet below.

However, even if a gain is specified, we are unable to provide specification limits outside of the datasheet, as this now becomes an issue of liability. Due to the random nature of noise, we cannot definitively say that all INA849 devices will have less than 0.2uVpp noise RTI.

However, since noise is a random behavior, we can use the probability of an event happening to our advantage here. If the customer is using the INA849 in a gain of 1000 and are expecting a p2p noise less than 200nVpp, we can calculate the expected noise. From the following TIPL video on noise (TIPL Noise Video), we obtain the below equations. Please disregard the numbers in the photo, they are from an example unrelated to our math.

Our normalized voltage noise is the voltage noise spectral density at a known frequency in the 1/f region. I have selected 1Hz, and at a gain of 1000, the noise spectral density is 3.8nV/rtHz. This gives us 3.8nV/rtHz * sqrt(1Hz) or 3.8nV. Multiplying this by the sqrt(ln(10/0.1)) we get 8.15nVrms. To translate noise in RMS into p2p noise. Since noise is centered on 0, the RMS noise can also be considered the 1 sigma noise. Typically the estimate that is used is 6sigma, which gives us 48.9uVpp, which aligns well with the typical graph for G1000 above. To achieve a peak to peak noise of 200nV, this would have to be a 24.539 sigma event. Assuming the devices follow a typical distribution, this will happen for 1 unit in 6.06x10²⁵¹ which is statistically very unlikely.

This does not account for events that fall outside of the standard deviation, like popcorn noise. Popcorn noise is a current noise, and will dominate with sufficient input impedance, as this current noise will now become a voltage noise and be gained up by the instrumentation amplifier gain. While TI does screen for popcorn noise, it is not possible to remove 100% of all popcorn noise occurrences. We test within a certain time window length, and if an event happens outside of that window length, the behavior will not be captured. Since these events can be an indeterminate length of time apart, we cannot say there will be no popcorn noise, as we will need to test for an infinite amount of time.

Best,
Gerasimos

Machida-san,

No worries on the response time! You are correct, this is a typical value, TI generally does not provide maximum specifications for noise.

Correct, due to the statistical nature of noise, we cannot provide a maximum specification, because noise can theoretically be infinite if measured over an infinite amount of time. However, the typical value provides a statistically reasonable expectation of noise, and further statistical estimations can be made to create more statistical definitions of noise.

Q2: Yes, this is mistake, this should be 48.9nVpp, not uV. I apologize for the inaccuracy, and thank you for your understanding!

Best,
Gerasimos