We have detected your current browser version is not the latest one. Xilinx.com uses the latest web technologies to bring you the best online experience possible. Please upgrade to a Xilinx.com supported browser:Chrome, Firefox, Internet Explorer 11, Safari. Thank you!

cancel
Showing results for
Did you mean:

# Adam Taylor’s MicroZed Chronicles, Part 110: SDSoC Coherency and FFT size

Xilinx Employee
0 0 43.4K

As we begin to use the Zynq SoC’s XADC, I first want to discuss two topics: FFT sizing and coherent sampling when sampling sine waves. Understanding these concepts is important to obtaining the most accurate information when we examine the frequency spectrum. They will allow us to examine the effects of our signal processing algorithms, best examined within the frequency domain.

Of course the XADC samples within the time domain so we must employ post processing to convert this time-domain information into the frequency domain. A Fast Fourier Transform is the most common method for accomplishing this task (see Xcell issue 91).

The theoretical noise floor for an ADC is given by the equation:

Where n is the number of bits of the ADC.

The FFT noise floor is determined by a similar equation, with an additional parameter to allow for theft processing gain:

The Zynq SoC’s XADC has 12-bit resolution and I am going to use an FFT size of 1024, which results in a noise floor of -101 dB.

With the FFT size determined, we need to ensure that we sample correctly using this FFT Size. At the most basic level, this is the difference between coherent and non-coherent sampling. If we sample coherently, there are an integer number of sinewaves within the data-capture window (which is also the FFT Size). For example if we wish to store 1024 samples we should ensure there are an integer number of cycles contained within that sample buffer.

Failure to ensure that there’s an integer number of samples in the sample window results in spectral leakage in the frequency domain due to discontinuities and these discontinuities impact the resultant frequency spectrum.

We can use the relationship below to ensure that we have an integer number of cycles:

Where Fin is the sampled input signal frequency and Fs is the sample rate.

We can rearrange this equation to correctly select an Fin that results in an integer number of cycles for the sample size and number of samples required:

It is best to use a prime number of cycles to prevent repetitive patterns and to ensure a random quantization of the noise floor.

The XADC has an Fs of 961538Hz, so if we want seven integer cycles and have a 1024-sample buffer, we require an input frequency of 6573.017Hz. Plotting this in Microsoft Excel, we see the following in the time and frequency domains. (Note: I have used Excel to generate all example data here not hardware.)

Seven Complete Cycles within the 1024-sample buffer

Coherent FFT only spectral content is within the bin of Fin only (6573.017 Hz)

Modifying the model to provide a non-coherent sample, we see the effects on the time domain and frequency domain below:

3.123 Cycles within the 1024-sample buffer

FFT of incomplete cycles showing smearing around the fundamental

We can use post processing and a Hanning window on a non-coherent sample to correct for spectral leakage. However, it is better to correctly consider the elements first.

Now we understand how to get the best performance by correctly configuring input signal and output post processing.

If you want E book or hardback versions of previous MicroZed chronicle blogs, you can get them below.

• First Year E Book here
• First Year Hardback here.

• Second Year E Book here
• Second Year Hardback here

You also can find links to all the previous MicroZed Chronicles blogs on my own Web site, here.

Tags (2)
Latest Articles