Tuesday, 29 August 2023: 2:45 PM
Great Lakes A (Hyatt Regency Minneapolis)
Solid-state radars employ pulse-compression: a long pulse (usually about 100 us, or 15 km in range in radar terms) is transmitted, which is modulated by some sequence. Upon reception of echoes from the environment, this specific sequence is searched for in the data using correlation (also known as matched - or intentionally mismatched - filtering). This approach, also known in telecommunications as "spread spectrum", is able to recover the signal from below the noise level, while also providing reasonable range resolution. This is needed in solid-state radars, because compared to magnetron or klystron radars, solid-state radars usually have one to two orders of magnitude less peak power.
In weather radars, usually a monotonic non-linear frequency ramp is used as the pulse modulation. The very exact shape of the non-linear modulation function (i.e., the instantaneous frequency) is crucial for the overall system performance.
Two main problems arise and are still prevalent at least in public opinion about solid-state radars, though they have been solved by the described pulse-compression waveform engineering: the first problem are range sidelobes - the modulating sequence is self-similar after a certain time shift, causing artifacts around strong point targets - and the second is a problem of blind range - as the modulated pulse is 15 km "long", the receiver does not "sees" the full pulse for targets closer than this distance.
In Meteopress, we have designed solid-state radar from ground up, having to deal with issues including the optimal waveform design. We employ an expert-engineered shape as a starting point for a genetic optimization algorithm. The radar is then pointed at a bright far-away target and the generated candidate waveforms are evaluated by transmitting them with the radar and computing a real-world correlation function. The score (fitness, loss, cost function) of each candidate is evaluated by comparing the measured correlation with an expert-specified "ideal" target, balancing main lobe width (range resolution), near- and far side-lobe levels, bandwidth usage and other parameters. The best candidates are then combined and mutated by the genetic algorithm, producing the next generation, and the process iterates.
Other factors that need to be taken into account are Doppler resistance of the engineered signal (Doppler shift usually worsens the sidelobes), the performance in near targets in the "blind range" where only part of the pulse is received, and also various non-linear characteristics, distortion effects in the radar chain, amplifier droop and filters with a non-flat frequency response, which damage the finely crafted pulse. This can be approached by digital predistortion and postdistortion at various processing stages.
In weather radars, usually a monotonic non-linear frequency ramp is used as the pulse modulation. The very exact shape of the non-linear modulation function (i.e., the instantaneous frequency) is crucial for the overall system performance.
Two main problems arise and are still prevalent at least in public opinion about solid-state radars, though they have been solved by the described pulse-compression waveform engineering: the first problem are range sidelobes - the modulating sequence is self-similar after a certain time shift, causing artifacts around strong point targets - and the second is a problem of blind range - as the modulated pulse is 15 km "long", the receiver does not "sees" the full pulse for targets closer than this distance.
In Meteopress, we have designed solid-state radar from ground up, having to deal with issues including the optimal waveform design. We employ an expert-engineered shape as a starting point for a genetic optimization algorithm. The radar is then pointed at a bright far-away target and the generated candidate waveforms are evaluated by transmitting them with the radar and computing a real-world correlation function. The score (fitness, loss, cost function) of each candidate is evaluated by comparing the measured correlation with an expert-specified "ideal" target, balancing main lobe width (range resolution), near- and far side-lobe levels, bandwidth usage and other parameters. The best candidates are then combined and mutated by the genetic algorithm, producing the next generation, and the process iterates.
Other factors that need to be taken into account are Doppler resistance of the engineered signal (Doppler shift usually worsens the sidelobes), the performance in near targets in the "blind range" where only part of the pulse is received, and also various non-linear characteristics, distortion effects in the radar chain, amplifier droop and filters with a non-flat frequency response, which damage the finely crafted pulse. This can be approached by digital predistortion and postdistortion at various processing stages.

