DRDO developed Arudhra AESA radar inducted in IAF
- Thread starter Laughing_soldier
- Start date
amardeep mishra
FULL MEMBER
- Joined
- Mar 8, 2012
- Messages
- 1,323
- Reaction score
- 26
- Country
- Location
@Oscar
When i wrote my comments earlier,i was purely pointing to the theoretical aspect of how signal processing is done in ordinary radars,however now i would shed some light on how exactly it is done on xilinc/alter's FPGA's with decent dynamic range of (16- 24bits)90-140dB
I was referring to a simple process known as "matched filtering"/"pulse compression" when i wrote my comments on the earlier page,But any matched filtering operation in time domain using a FIR filter(it can very well be accomplished in frequency domain using good old FFT) is followed by doppler processing in frequency domain.
Suppose you're using sampling a PRF of 1kH with sampling frequency of 1MHZ,then you'd get 1,000 samples,Now these 1000 samples are from a single PRF.These are stored vertically in bins/registers as shown in the pictures below.Each vertical bin constitutes a single PRF sampled data that is matched filtered with transmitted waveform(as i explained in my posts in last page).Now matched filter is specifically used because it "maximizes SNR" .The output of matched filter would be 1 or high only when output of this cross correlation operation exceeds a certain threshold(you can see the maxima occuring in the red graph shown in second figure) and if that occurs,then the entire contents of that bin are loaded vertically in another bin(note till now the data is still in time domain,however this very same process can be also implemented using FFT instead of a FIR filter)
In doppler processing though, the data is read horizontally from vertical columns(obtained from the previous step) and FFT is performed over that sampled sequence by the very famous equation y(n)= summation(x(k)*exp{(-2*pi/N)*n*k}) where k varies from 0 to N-1.Here i would like to bring your attention to the fact that the equation i have written above is that of DFT and that FFT is implemened using various known algorithms like "cooley-tukey" algorithm etc on a FPGA. FFT is obviously preferred because in 1024 point data set,FFT takes just 1% of computational requirements of a DFT,this is more significant when the point of data set increases
@Oscar
pulse compression or chirping is another very interesting phenomenon all by itself! -a good way to improve range resolution
When i wrote my comments earlier,i was purely pointing to the theoretical aspect of how signal processing is done in ordinary radars,however now i would shed some light on how exactly it is done on xilinc/alter's FPGA's with decent dynamic range of (16- 24bits)90-140dB
I was referring to a simple process known as "matched filtering"/"pulse compression" when i wrote my comments on the earlier page,But any matched filtering operation in time domain using a FIR filter(it can very well be accomplished in frequency domain using good old FFT) is followed by doppler processing in frequency domain.
Suppose you're using sampling a PRF of 1kH with sampling frequency of 1MHZ,then you'd get 1,000 samples,Now these 1000 samples are from a single PRF.These are stored vertically in bins/registers as shown in the pictures below.Each vertical bin constitutes a single PRF sampled data that is matched filtered with transmitted waveform(as i explained in my posts in last page).Now matched filter is specifically used because it "maximizes SNR" .The output of matched filter would be 1 or high only when output of this cross correlation operation exceeds a certain threshold(you can see the maxima occuring in the red graph shown in second figure) and if that occurs,then the entire contents of that bin are loaded vertically in another bin(note till now the data is still in time domain,however this very same process can be also implemented using FFT instead of a FIR filter)
In doppler processing though, the data is read horizontally from vertical columns(obtained from the previous step) and FFT is performed over that sampled sequence by the very famous equation y(n)= summation(x(k)*exp{(-2*pi/N)*n*k}) where k varies from 0 to N-1.Here i would like to bring your attention to the fact that the equation i have written above is that of DFT and that FFT is implemened using various known algorithms like "cooley-tukey" algorithm etc on a FPGA. FFT is obviously preferred because in 1024 point data set,FFT takes just 1% of computational requirements of a DFT,this is more significant when the point of data set increases
@Oscar
pulse compression or chirping is another very interesting phenomenon all by itself! -a good way to improve range resolution
Last edited:
amardeep mishra
FULL MEMBER
- Joined
- Mar 8, 2012
- Messages
- 1,323
- Reaction score
- 26
- Country
- Location
crow of data on which FFT is performed belongs to the sampled data at ith instant across N PRFs,this is taken over CPI(coherent pulse interval).The time interval over which pulses remain coherant
