Using an assortment of microwave transmitters, receivers and transceivers (operating at frequencies of 10 GHz, 25 GHz, and sweepable 10 GHz to 30 GHz) the system will basically work as follows:  

There will be one emitter located at the rear center--near the CHMSL--and it will have a rather broad beamwidth in the horizontal plane (covers three lanes).  

There will be four receivers (having narrower beamwidth), two mounted on each side of the vehicle--one high, one low--both aimed straight back. (I am using only one on a side for now.)  
  
  Three modes: 
1) Mode 1, Doppler:  
The Doppler mode will sense the speed of the vehicle-to-be-warned, and the speeds of the various rear-approaching vehicles. To enable mode-1, the emitter is made to (switched) "leak" some of its carrier RF--in equal amounts--to all receivers, thus producing Doppler outputs (unique to each) from each receiver. It should be noted: these receivers will be able to differentiate between approaching and receding traffic.  

2) Mode 2, Differential:  
The relative reflected intensities detected by the four receivers should help in the discrimination of the various vehicles' locations. This second mode, will not have the leaky carrier, but will detect relative CW signal strengths from each receiver. By analyzing the relative differences between what each antenna "sees," there should be an enhancement of the positional information. This coupled with range and velocity data, should be synergistic.  

3) Mode 3, Range:  
The Range mode is used to derive distance information of the target vehicles. This mode will also not use leaky carrier.  This method of range or distance measurement uses a relatively low bandwith phase nulling approach, not fast, wide bandwith pulse techniques. It will use swept AM modulation, i.e., the emitter (Gunn oscillator & antenna) will be AM modulated with a linearly swept sinusoid, having a frequency ranging from ~ 1 MHz to ~ 10 MHz.  

Principle:  
At a distance = 1/4 wavelength of the modulating sinusoid, the reflected carrier at the receiver--after demodulation--will be 180 degrees out of phase (null) from the originating modulating wave. Knowing the exact frequency of the modulating wave, the approximate distance to the target can be derived. (remember, 1/4 wavelength is the round-trip distance for 180 degrees or1/2 wavelength) Note, there will be some error in this approach if the rate of sweep--of the modulating wave, verses the range isn't accounted for.  

As the modulating frequency continues to increase, the 180 degree "null" will again occur when the modulating frequency is 1.5 times the first null frequency (f_m); this will continue to occur every additional half wavelength, i.e., 1.5 f_m, 3.5 f_m, 5.5 f_m, 7.5 f_m, etc.--each time resolving the distance ever finer. That is: 1/2 WL at 10 MHz is a finer unit of measurement than 1/2 WL at 1 MHz.  

All three modes will be used in combination in a very rapid fashion. (limited by PRF, Doppler frequencies at n x MPH @ m x GHz, as well as, the sweep rate of the modulating frequencies, f_m)  

  Summary: 
If this approach proves to have merit, then I propose that the next version use some sort of pulsed, coded spread spectrum, i.e., TDMA, CDMA, FDMA, etc. Each has different advantages and disadvantages,i.e., TDMA (Time Division Multiple Access) might be safer (from a health standpoint) because it emits RF energy less than full time; CDMA offers better security/reliability; FDMA is better surviving local interference, etc. 

This is a daunting task: because of the high reliability needed, the operating environment, the kinds of information needed, the finite time allotted for its detection and interpretation, and the cost target as an automotive OEM device  

It is a given, that the success of this approach will depend, in great measure, on good signal processing (DSP) techniques.  

With an assortment of microwave transmitters, receivers and transceivers--10 GHz, 25 GHz, and a sweepable 10 GHz to 30 GHz--to test an idea I have.