back
91 comments
Shahed drones have increased in altitude from ~500m at the beginning of the Ukraine war to 2000-3000m, which is a 4x reduction in noise on the ground. The higher the drones are, the less noise they make at ground level, and the less effective this ground-based microphone system will be. The drones have moved to elevations to make them more difficult to target with ground based weapons. Reductions in ground noise are a secondary effect.

The latest versions of Shahed can reach 5000m in altitude, which would largely be inaudible on the ground.

At those higher altitudes they are trivial for radar to detect, and from farther away too. They are adapting cheap commodity used marine radars to get this done in some places

I suspect the Shaheds are going higher to mitigate AA ground fire. Higher up you have to send a missile or interceptor up.

It’s a trade off.

And interceptors are more expensive than Shaheds. A $100k interceptor (guessing here) to shoot a $10k Shahed can be an acceptable deal for the side launching the Shaheds.
Ukrainian P1-Sun interceptor FPVs were built specifically for this purpose. They are way cheaper than a Shaheed drone and have a practical ceiling of 5000+ meters according to wikipedia.

https://uk.wikipedia.org/wiki/P1-Sun

It's not about "can reach", it's fairly easy to get them to fly even higher. It's about danger of interceptors vs danger of detection. Today's Ukrainian detection network (based on radars) is so dense there is no way to hide from it anywhere, anyway, so high altitude wins.
> It's about danger of interceptors vs danger of detection.

What's with that "vs" trade-off?

You're saying avoiding detection requires high altitudes.

What do interceptors have to do with that?

> ~500m .. to 2000-3000m, which is a 4x reduction in noise on the ground.

Radiative power drops by the square of the distance? Does anyone with a real physics background start demanding authority on a soapbox like this?

Could you simply use small cheap balloons with microphones to listen in higher altitudes? Or even small drones that carry the microphones with their own engine noise masked out?

Edit: as others say, plain radar will suffice.

So film the sky during charging and run a llm on it?
Or an image detection model. Fraction of the compute and can run even on edge embedded. And easy to train with your own data
Clouds and nighttime are a barrier to visual detection. Even with good effectiveness the conditions needed for that would mean that you have far less than 50% uptime, and your downtime is predictable to your adversary.

A cheap radar takes an order of magnitude less power to run on hardware that is cheaper than an LLM and can see way farther than a camera.

Language models are terrible at computer vision. I don't know why you think that a software to generate words would be able to see flying objects in a picture.
Or it is a public stunt? I am naysayer engineer. Build acoustic detection system for anomaly detection in manufacturing line. It worked with studio level equipment with lots of noise filtering measures. First challenge was to have microphone listening to the right place and getting reliable signal. Second challenge was to amplify the right signal to a level where the features could be extracted. I doubt an old android with dirty microphone can detect anything reliably.
If these Shahed drones have a propeller they should have a brutal signature between 0-250Hz as they are moving a lot of air. And if the engine speed of the Shahed stays constant it will be even easier to detect it ("Angry lawnmower sound".. but that's only the high frequency part of it).

There are extremely sensitive differential pressure sensors (like SDP600-25Pa) available from Sensirion that aren't overly expensive.

Use one differential side and connect it to a kitchen funnel for directional listening the other one to a plastic bottle with a tiny hole in it. This way the sensor will "Null" out the environmental pressure (which the bottle follows very very slowly) from both inputs. It then only will pick up everything high frequency which is left over (and the bottle cannot follow because of its small hole).

This way I was able to detect washing machines that have a physical link to a house from many hundred meters (machine spinning -> house wall shaking -> pressure waves) away. The speed pattern of washing machines when spinning is very unique (several steps over many seconds).

Add this with some GPS PPS frame timestamping and you should have a nice tracking network that doesn't require a lot of bandwidth. But maybe the setup must switch to analog differential pressure sensors as these Senirion-I2C sensors do not have a Sync ping for super precise timestamping.

> Use one differential side and connect it to a kitchen funnel for directional listening the other one to a plastic bottle with a very very small hole in it. It will pick up everything high frequency that's different to the environmental pressure.

Nevermind drones, and war, that's all fine; but I need to know more about this. Is there a phrase or name for this I could use to find more information, maybe example schematics?

The Sensirion SDP600-25Pa speaks I2C and only has a handful of primitive commands. Add this and 3.3V and you are done.

I'm 99.5% sure if you throw Claude with a datasheet on it will Slop out working code for a ESP32 with ESP-IDF.

Interesting! There are a lot of super loud cars in my urban area. I want to catalog where and when they drive, so I can stand on the side of the street and shake my cane at them.
I’ve always dreamt of building a mesh network of loud vehicle detectors that would map current loud noises and track where loud vehicles live, coupled with very powerful parametric speakers. So when e.g. the network detects a loud motorcycle at 3am anywhere in town, that live sound is instantly played at high volumes targeting only homes where other loud vehicles live.

Somehow there has been little VC interest in this idea.

Isn't detecting pressure waves in air exactly what microphones do?
Yes. But they usually are not performing very good between 1-250Hz.

Sensirion is using a thermal flow-sensing principle method which is basically a heated plate that cools/heats up when air passes it - making it extremely sensitive in this range.

The equivalent noise density of the sdp600 is 10^-2 Pa/sqrt(Hz). While the noise density of a typical microphone is 10^-6 Pa/sqrt(Hz). At same 50hz low frequencies. So either you're full of shit and a high quality microphone is x10000 more sensitive than this sdp600 sensor, or I'm not understanding what you're saying.
Any self-respecting microphone will be sensitive to tens of micropascals (the conversion between SPL and Pa is straightforward). The Sensirion device is proud of its ability to operate at 10 Pa.

That being said, most microphones are intended for audio and their amplifiers and readout circuitry are not optimized for very low frequencies. And the physical measurement devices don’t work down to DC. (If they did then the microphone wouldn’t work — the DC pressure is around 1 atm, which is some 12 orders of magnitude above the more quiet sounds the microphone is trying to detect.)

A device sold as a pressure sensor will be intended for use at or near 0 Hz and will work fine.

I assume the actual practical crossover is something like 0.1 - 10 Hz depending on the microphone. There’s an additional practical issue, though: the pressure sensor comes from the factory already calibrated for DC, whereas you may be on your own trying to get physical units out of a microphone much below 20Hz.

If you use a high end microphone intended for professional use with phantom power, you will also be fighting the power circuit and lower frequencies.

Nice. And instead of smartphones, dedicated sensor using esp32 makes more sense imo.
What’s the frequency response of that sensor? The datasheet says nothing.
Honestly: don't know.

My experiments never had a dependency on linearity.

Interesting! I will like to see your circuits, if any.
Not much circuits needed.

The SDP600-25Pa speaks I2C and only has a handful of commands.

Just read it out with a microcontroller you love (like ESP32) and send the samples to a host for analysis. The ESP32 has limited I2C time stretching capabilities limiting it in the highest resolution modes of the sensor - but often that's not a big factor.

To not overwhelm the poor processor and Wifi maybe better a bunch of frames (like 512 or more).

Doesn't wind fool this?
Theoretically wind can be Nulled with a 4-way water splitter that are placed in 90° angles (+ a connector at the bottom to the sensor). As long they are orthogonal the wind pressure coming from one side will cause a negative pressure on the opposite.

Only pressure waves coming into all holes at the same time will reach sensor.

Never tested it. Only a Gedankenexperiment as Einstein would say.

I wonder how this system can be protected from spam - if anyone can send data there, enemy can, too
I feel like if/once they reach the number of expected participants (10k), it'd be easy to filter out the spam as long as the majority are truthful.
So what prevents enemy from creating 20knparticipants constantly sending fake data?
Sounds like a simple app with mic input being sent to a yamnet-like audio classification model for a single target detection. Hardly anything innovative?
For those interested there is an open source radar based phased array project here: https://github.com/NawfalMotii79/PLFM_RADAR

I may have even discovered it from hacker news, I forget now.

I remember seeing this on HN when it was first submitted. I honestly think that project is either a scam, an elaborate hoax, a cover for a Chinese company selling radar systems or a three-letter-agency honeypot.

The files [1], [2], [3] on Github (once featured on README and on some articles) depict what I presume is a Chinese built radar which you can see here [4], [5]. I also suspect this might be a honeypot since Instagram - of all places - is full of accounts offering these radars [6], [7]. Some of the contributors on that projects are heavily leaning into AI. Git project history is suspicious. Discussions on the issues seem very artificial. I remember having bunch of notes on this project, but can't find it now. Just looking at it from outside, nothing about it seems real.

[1] https://github.com/NawfalMotii79/PLFM_RADAR/blob/main/docs/a...

[2] https://github.com/NawfalMotii79/PLFM_RADAR/blob/main/docs/a...

[3] https://github.com/NawfalMotii79/PLFM_RADAR/blob/main/docs/a...

[4] https://www.cloudwalkerfpv.com/product/CW-t20.html

[5] https://www.militarysignaljammer.com/supplier-4701497-survei...

[6] https://www.instagram.com/cloudwalker_radar/

[7] https://www.instagram.com/radar_lisa88/

While you may be correct in some of these assumptions, the repo is open source, contains actual schematics and code, appears to have real community contributions, and also happens to exists now, which is something that can't be said for the original post this was a response to. I was hoping to just provide some reference to existing work that I found interesting since it is what I think is a much more traditional type of detection system.
Wouldn’t a purpose built Esp32 with microphones aimed at the sky do a better job? It would be always on, better directional targeting.
I believe Ukrainians had already deployed such a system. This is specifically designed to use old Android phones already sitting in a drawer somewhere without any other use, and most importantly by anyone without technical skills.
AFAIK started by a fellow HN’er
[flagged]
As much as I like to point and guffaw at "bad evil rocket man" the from orbit bit is doing a massive amount of lift (pun not intended) for that price.
Multi billion dollar spend to “see” $2,000 aircraft and try to shoot it down with multi million dollar equipment. The speed to produce and cost effectiveness of these drones are seemingly at a massive advantage right now vs Defense capabilities and straight up cost and looks to be that way for a good while.