If you have 9 antennas and 599 samples you cannot have the same number of samples from every antenna. Why not? For most standard algorithms that would be essential.

Are the 9 antennas in regular rectangular array (3x3)?

What is the width of the array in wavelengths at the highest and lowest frequency?

What is the angular separation of the two targets from the centre of the array?

What angle does the array present at the targets (how wide does it look in degrees)?

You can only reliably separate targets that are one beamwidth apart at the highest frequency. This is roughly 52 degrees times wavelength/aperture.

Your antenna separation shouldn't really be more than a wavelength at 2 GHz, so 150mm, and the array might be 400 mm square. This gives a resolution of 13 degrees at 3 GHz so if the targets are closer together than this you won't separate them using FFT methods, although maximum entropy or maximum likelihood methods might.

You can also process the data yourself by designing a phase removal or focusing algorithm. This is what the FFT does, but it requires regular data samples and, probably, targets in the far-field.

For each possible target position, work out how far it is from each antenna, and for every data sample, at each frequency, work out the phase from the antenna. Then subtract the phase from the signal from that antenna, so that all antennas should illuminate a target at that position at the same phase. Then add the signals from all the antennas together. Do this for all lots of possible positions of the target, and if there is a target there it should focus and give a peak, and if not, the signals from a target at another position should not be in phase and won't add up to so much.

To make sure you don't miss the targets, choose positions separated by the resolution or a bit less.

This is what the FFT does, but much more efficiently that the method I have described, but my method should work for random positions of antennas and random frequency steps, and in the near field.

There are methods to get FFT to work in the near-field, such as the STOLT algorithm, if your array has regular steps in space and frequency. I don't know why it isn't working for you, you have not given anywhere near enough information.

I understand your doubt. But the 599 samples are for each antenna creating a Data frame of 9 columns and 599 rows

OK, that makes sense. The processing method I describe should still work. Matlab is designed to make that sort of processing more straightforward.

Yes can you please describe how?. And Actually can Crossrange and DOA calculations might look same at certain point

Do u have any data from real radar?? Actually i am testing my prototype algorithm for DoA estimation.