FFT IP not behaving as expected.

[Sean2]

I'm trying to use the FFT IP on Quartus Prime Pro 18.1 to do an inverse fft. When I input an fft of a signal I created in Matlab that has a frequency of 23, I do not get the expected output signal. When taking the fft, with Matlab, of the output signal, I get a frequency of 86.

Before I input the signal to the inverse FFT IP, I make sure to convert the signal to Radix-4 digit reverse order with the Matlab code provided in the FFT IP data sheet.

I tried doing an fftshift and inverse fftshift on both the input and the output to see if one was shifted, and this not fix the issue.  

[SengKok_L_Intel]

Is the input data compliance with the single floating-point format? Does it work well with FFT before proceeding with iFFT? How about if change the input order to natural?

[Sean2]

The input data is in single floating-point format. When I change the input setting of the IP to natural input and input the signal in it's normal order, I get a very noisy output that is not at the desired frequency. When I use this natural input setting and input the data in radix-4 digit reverse order, I get a frequency spike at 86 instead of 23.

[Sean2]

When I use a testbench with just an FFT IP instead of an inverse FFT, I get the expected output of a spike at 23 after using Matlab code on the output to rearrange the radix-4 digit reverse order. However, when I hook up the FFT IP to the inverse FFT IP through Avalon streaming, the fft produces a different output. 

[SengKok_L_Intel]

Do you able to get the expected result if running in the RTL simulation?

[SengKok_L_Intel]

If further support is needed in this thread, please post a response within 15 days. After 15 days, this thread will be transitioned to community support. The community users will be able to help you with your follow-up questions.