I need to design a very short range pulsed radar for classroom/lab use. For this purpose the pulse width needs to be very short 1-5 ns to avoid the radar blind range. My preference is to digitize the received pulse and do all the processing inside a general purpose processor. But for such a short pulse width radar, I need a very high sampling rate i.e., at least 2 G samples/sec. Digitizers of this rate are very expensive and cannot be afforded in the budget. Can you guys suggest any tips/tricks to detect such a short pulse width at low cost?

• And how exactly do you envision the general purpose processing at 2 Gs/s rate? Even if you have the digitizer? – Ale..chenski Apr 20 '18 at 5:48
• Thats a good question. I was assuming we could somehow do threshold triggering where we get data from only valid pulses. this will reduce the processing requirements – Faisal Apr 20 '18 at 6:51
• @AliChen pulse radars don't have to sustain that rate - you capture to some device-internal RAM, then stream at whatever the (USB/Ethernet/PCIe…) link allows to the host computer, then effectively process offline – Marcus Müller Apr 20 '18 at 6:51
• @MarcusMüller, then I re-direct the question to you: how do you envision to capture the pulse to "some device-internal RAM" at multi-GHz rate without specialized high-speed HARDWARE? (which costs thousands in development, requires deep engineering expertise in ultra-high-speed electronics, and is clearly way above the "classroom/lab" student level). That was my whole point, about the staggering disconnect between Arduino/Raspberry level and the reality of radars. – Ale..chenski Apr 20 '18 at 19:05
• Definitely agreeing with you here. If you really need to do the whole signal proc in digital domain, you'll start with a 10 k$FPGA dev board and something slightly cheaper in ADC hardware. – Marcus Müller Apr 20 '18 at 19:07 ## 3 Answers I'm not sure if your goal is to do the design your self or to just arrive at the functional unit within budget. Regardless here is a link to a pulse radar MMIC sensor that operates at 39 G samples/s. They offer multiple dev kits with some coming in under$300. You can choose to provide the software/MCU yourself or choose a kit where that is already done as well.

What signal level to expect. With 10GigaHertz bandwidth at Rnoise of 62 ohms, the noise floor will be 1 nanoVolt *sqrt(10^10 Hz) = 1nV *100,000 = 100uVolts. For 20dB SNR, you need 1milliVolt from the antenna, or (-)56dBm.

Assuming you know the carrier frequency, mix that with what comes from the antenna. Amplify the mixer output (which should be baseband, having been homodyned) by 56dB; thus your signal level will be at least 1 volt. Feed this signal into a number of coaxial delay lines, all but one into a diode-bridge sampler. The one non-sampled channel feeds an analog comparator, which trips all the diode-bridge samples concurrently.

You will need to trim the delay-line lengths.

Your major cost may be the coax cables. Enjoy.

Rather than trying to capture a single pulse, what you want to do is send out many pulses at a fixed rate. Let's call this the pulse repetition rate, or PRF. The PRF could be something like 10 kHz.

Since the TX pulses are the same and the target is not changing, the receive waveform, whatever it is, will not change from one pulse to the next. So you can sample at a frequency just a tiny bit slower than the PRF. If the period of the PRF is 100us, you can set your sample period precisely to 100.001us. So each subsequent sample will come from a slightly later point on the receive waveform. This has the same effect as sampling a single receive pulse at 1GHz. As long as the TX waveforms are identical, and the target doesn't move much from pulse to pulse, this technique is valid.

It requires precise clocks and frequencies and whatnot. But I believe that can be achieved using DDS IC's running off of high-quality frequency sources. If you need to sample at an effective 2GHz rate, that means you want the PRF period to be 100us, and the sample period to be 100.0005us. Hopefully you get my drift. As long as the sample frequency is just a tiny bit slower than the PRF, the difference between the pulse repetition period and the actual sample period becomes the EFFECTIVE sample period.

• May still require a very special ADC, because the ADC analog bandwidth still needs to be high enough for the raw RX waveform. – mkeith Apr 21 '18 at 7:28