SINAD, THD and sampling rate

[rsv]

I am doing noise measurements on an ADC. We use an input signal of 1 kHz, sample at 11.025 kHz, and measure THD and SINAD. We are comparing our values to the values that are listed on the datasheet for the ADC, and there is a large difference (around 30 dB). The datasheet says the input signal they used is also a 1 kHz sine wave, and they are sampling at 48 kHz. Is it possible that this difference in values is due to the difference in sampling rates?

[LV_Pro]

I discuss some of the issues in another thread : http://forums.ni.com/ni/board/message?board.id=170&message.id=124898

I'm not sure that you should be seeing 30 dB difference, but yes, the sampling rate may be part of it, and how many samples you are evaluating. If you have a steady state sine wave, and aren't looking for fairly quick transients, the longer that you can sample (the more samples evaluated) the better. As the number of samples goes up so does the resolution of your FFT (the bins get narrower), meaning that the average energy in each bin is more representative of your actual signal. Are you using any windowing of the data, filtering of the input?

P.M.

[rsv]

We are using a Hanning window, but we are not filtering. Could a different window or filter help improve the accuracy?

Thanks for the help.

[LV_Pro]

It might, but the most immediate thing to look at is the sampling rate/number of samples. Also, when you say that there is a 30dB difference, are you sure that you are using the same units. How many samples are you sending to your FFT?


P.M.

[rsv]

We are sending about 11000 samples. Is there a way to calculate what the sample size should be (an "ideal" sample size)? Also, is there any benefit to using a filter (such as a Butterworth) for calculating THD?

Message Edited by rsv on 06-09-2005 02:03 PM

[johnsold]

FFT algorithms are most efficient when the number of input samples is an integer power of two. 8192 or 16384 are the closest powers of two to your 11000 samples. A zero padding VI is in one of the signal analysis palettes if I recall correctly. It appends zero-valued samples to the end of your data array to make the fianl array have a power of two length.

Lynn