The only way to secure against the described exploit is to measure round-trip-time from the car -> key -> car and ensure it's under, say 5 light-meters: aka 16 nanoseconds, plus the carefully calibrated time it takes the key to compute its response.
16 nanos is a very short amount of time, and it'll be tricky to measure that reasonably accurately.
The real solution is to require the user to interact with the key in some way, like pressing a button, or perhaps moving it around (as would happen as you walked with it in your pocket).
One attack vector is stealing cars out of a supermarket parking lot. You just wait for someone to drive up in the model car of your choice and have your accomplice discreetly follow them into the store. When your accomplice texts you that they are at the bread aisle/back of the store you can just steal with impunity knowing that a bystander will see no difference between the actual owner who has the key in his pocket and you with your relay device in your pocket. You also know that your victim is in the back of the store and that they can't get within sight of you before you're already long gone.
>A PKES car key uses an LF RFID tag that provides short range communication (within 1-2 m in active and a few centimeters in passive mode) and a fully-fledged UHF transceiver for longer range communication (within 10 to 100 m). The LF channel is used to detect if the key fob is within regions Inside and Outside of the car. Figure 2(b) shows the areas in proximity of the car that must be detected in order to allow a safe and convenient use of the PKES system. The regions are as follows. [1]
1. http://www.syssec.ethz.ch/content/dam/ethz/special-interest/...
Yes, the actual key itself is located by the car based on Low Frequency RFID.
The attack described is a relay attack, which means that the key can be spoofed in real time by relaying short range radio transmissions to two locations.
The mistaken assumption of the security system is that the short range communication protocol used by the car and the key requires the key to be in close proximity to the car.
Since the communication may be relayed, the range assumption is invalid. The main suggestion is to use high precision timing to determine the range, as it is very difficult to cheat on the speed of light.
I agree that "signal strength" is not the best way to phrase the above in a technical discussion.
I have not seen any indication that triangulation or any other physical location system is used in vehicle PKES.
Once in pairing mode, the physical key and ECU use standard public-key crypto (ala SSL) to setup a secure connection, then exchange keys.
In theory you could allow boot-strapping another key so long as an existing paired key is present which would make the procedure above your failsafe for when all keys are lost/destroyed. If you wanted to take things a step further you could use a form of distributed Kerberos where the manufacturer sets up a physical key with a ticket allowing access to one (or a set) of allowed cars but that makes the manufacturer's systems a massive target for hacks/social engineering which is a problem because thousands of dealer technicians need access to those systems... that's the point of the delays and short acceptance windows above. An evil tech or hacker can't pre-create a bunch of keys on the sly.
To unlock or remote start, the key broadcasts a HELLO message, encrypted with the ECU's public key. The ECU responds with an ACK+nonce encrypted with the physical key's public key. The physical key decrypts it and replies with an ACK+nonce encrypted with the ECU's public key. Congrats, you now have a reasonably secure system that prevents replay attacks.
Ultimately it would require embedded software engineers and company management who a) understood security and b) gave a shit. Both are in extremely short supply.