- May 18, 2018
- Reaction score
^^^ This ^^^^I'm not really a countermeasures enthusiast (yet), but I came across this project the other day as a suggested youtube video. Really cool stuff you are doing here in SW Ohio, Jon and Co.!
A thought I had; you could use similar technology for lidar jamming. Specifically, while you've used a CNN to train for classification, it would be possible to use something like NEAT (NeuroEvolution of Augmenting Topologies) on top of a MLP to train a lidar jamming strategy. So you have a CNN classifier on the receive end, and for the response, all you'd have to do is pipe the error state of the gun back into the "response" network as an input for training. While a conventional NN takes advantage of gradient descent on a fixed topology, which limits solutions to local minimums, NEAT allows both the weights and topology to evolve. Theoretically, you could create a network which evolves the most optimized solution to defeat any given lidar. The only catch is that you need the error state of the gun, basically in real-time, which means you need any gun you'd want to train for. Since there is a response involved, and the lidar's reaction to that response is the desired outcome, you couldn't rely on recoded data. Although this has to be true for any lidar jammer manufacturer, they're just likely manually experimenting with responses/jamming patterns until they find something that works, with perhaps some degree of reverse-engineering the internals. As you've mentioned in videos, the main benefit would be the faster time to market with better, more effective jamming strategies than those found through manual experimentation. You could probably even do this with a vanilla network, but I have to assume that lidar manufacturers are using some pretty sophisticated anti-jamming, and if not today, will be, and a genetic algorithm like NEAT could likely brute force an optimized solution for any current or future anti-jamming strategy. It's all just a beam of light, after all.
Just thinking out loud here. Carry on.
I had mentioned when Rai topic came out, that eventually Radenso could cut a deal to license some of their IP in the future for laser CM Partners (Or Rad create their own system). Assuming they stayed far ahead in this new order of thinking. Some of the above will be applied, just a matter of time. And will apply nicely to laser CM.
The above starts to explain what real AI is/does. Thank you for exposing some more of the reality. Welcome to RDF. As a former Oakwood (“under the dome” suburb of Dayton) citizen from the stork delivery to 22ish, happy to see Daytonion so well versed in the future. If you are a transplant, hit the Pine Club (if you have not already).