Piccolo Processor A to D performance

Hi

Below thread may provide some useful information about you query.
e2e.ti.com/.../21364

regards
Aditya

Thanks, seems to indicate that one board/processor has an RMS level of about 1 bit. I am looking for the max or typical expected value for any piccolo.

Keith K.

Hi Keith,

The SNR is approximately 66dB and the standard deviation of a DC input histogram is approximately 0.7LSBs. These are both typical values; we unfortunately don't have any more detailed measurements.

Can you send me the histogram data? That is what I am looking for. I don't see that in the data sheet.

Hi Keith,

We don't have raw data for the histogram.  

Interestingly, even with a Gaussian noise source on the input, the output may not look quite Gaussian due to the DNL of the ADC.  Here are 3 simulated graphs with randomly generated Gaussian noise with center code = 128, noise STDEV = 0.7, and no DNL errors - these look pretty Gaussian:

However, when we add some simulated random DNL errors, the graphs may looks quite a bit less Gaussian, even though the input noise is in fact Gaussian:

(Note that the DNL in this case is the average DNL, not the Min/Max DNL...these distortions wouldn't occur throughout the whole transfer function for a normal ADC, only at some worst case spots)

All of this is to say that I don't think a snapshot of the input histogram is all that useful; you are much better off working with the SNR, as this captures the noise across the full ADC range. 

Thanks, very informative. When I use SNR to calculate RMS noise I get 1.7 bits. But your saying it's more like 0.7, is my math right? If so what else can explain the difference?


SNR = Full Scale/noise = 4096/(quantization noise + other noise) = 66dB

66dB * (quantization noise) + 66dB * (other noise) = Full Scale

Other noise = Full Scale/(66dB) – quantization noise

Other noise = 4096/1995 – 1/sqrt(12)

Other noise = 1.764 counts.

Does that sound correct?

Hi Keith,

If I back calculate, I get 1.347 LSBs.  Some notes:

-When adding the noise sources, we need to add the noise power, not the RMS values directly (hence SNR = 10log() instead of 20log())

-SNR is typically characterized with a sine input, so the signal part of the ratio has an RMS of (2^N)/sqrt(8) instead of the (2^N)/sqrt(12) for a ramp that covers the full range.

Awesome thanks.  So which is the better number to use 0.7 or 1.3?

Keith K.

I would probably use the 1.3...as previously stated, measurements of input histogram may not exactly reflect the input referred noise due to DNL variations. What is your application?

3 phase brushless motor control loop in a torque generating application; motor is mostly sitting still. A to D is used to sample the output of the current sens amplifier. We are using FOC control. Our end product is sensitive to audio noise. Noise in the motor loop turns in to audio noise through the motor so we are trying to minimize all noise sources. We have cut the bandwidth of the output of the current sense amplifier but did not see much drop in the noise. I think my last measurement with DC input was 1.28 counts RMS. By calculation the band limited output of the current sense amplifier should be about a third of a count. I am wondering if there is any board layout issue that is coupling noise in. This would be something to chase. We are sampling faster than we are using the data so we implemented averaging of 4 A to D samples. This cut the std of the noise in half which correlates to 1/root(4). I think all this adds up to indicate there is little coupling of noise due to the board layout and we are doing the best we can on this front.

Thanks so much for your help.

Keith K.