Dense Traffic these documents, drawings and specifications are the property of roadeye flr general partnership, and shall not be reproduced or used without written permission from roadeye flr general partnership. RoadEye

3.5Signal Processing

A very large amount of effort has been invested in the development of a suite of algorithms which one hand are sophisticated enough to cope with the very complex situations observed in automotive scenarios, both in highways and in city, and on the other hand they are as simple as possible that can be developed, tested and proved reliable.

Simple algorithms can be developed based on models and simulations, but they will not be adequate in real situations. On the other hand it is impossible to predict and simulate the variety of situations in the real world. Therefore, the algorithm development at RoadEye has been performed, as it is usually done in the Radar community, through a constant cycle of data collection, algorithm development, testing on the data base, algorithm refinement, additional data collection, and so on.

The data collection setup basic requirement was the ability to collect the raw samples directly from the ADC (Analog to Digital Converters) for continuous periods of time of about 2 minutes together with a synchronized video picture of the road scenario. Several hundreds of Gigabyte of data was collected this way in Israel and several countries in Europe (Holland, Germany, Sweden…) in various traffic scenarios (fast highways, country roads, dense city and climatic conditions (hot, cold, very cold, light and strong rain, snow and strong snow, fog…). The data collected was essential to the algorithm development. It is impossible to imagine the development of a radar sensor for ACC application without this tool.

A
fter some two years or more of algorithm development, the level of false alarm became low enough that the original method of just recording and catching random events was not effective. We were generating too much no longer relevant data. We therefore developed an application that keeps a circular buffer of data and when some specific interesting situation happens, then we press a button and download to the hard disk some ten seconds before the specific event and record until stopped. This procedure allows the collection of data relevant to the further improvement of the algorithms.

The signal processing implemented is best described in the flow diagram shown in Fig. 11. The important difference with other existing FLR’s is that the transmitted waveform can be modulated at will with simple software commands. This freedom entails the development of a Waveform Selection Logic that should determine which of the waveforms will be transmitted and processed depending on the situation.

A
t the beginning of the development, we relied mainly on FMCW waveform transmition and processing. The FMCW principle of operation is shown in Fig. 12. We obtained rather good performance in Israel and in Holland, but after some data collection campaigns in Germany, we observed that the very strong returns from the poles that hold the guard rails hindered the target detection and distorted the angular measurement. In Fig. 13 a graphical explanation of this phenomenon which results from the lack of discrimination between moving targets and static objects.

We therefore decided to implement a Range-Doppler waveform to overcome this difficulty. The principle of operation of the Range-Doppler waveform is described in Fig. 14. This waveform has the Doppler discrimination capability (as shown in Fig. 15) and is widely used in defense radar systems but it is the first time that has been applied to FLR’s in automotive applications.

The implementation of the Range-Doppler technique included the design of several different waveforms with range and range resolutions appropriate to the varying scenarios. Each waveform was implemented with three different PRI’s (pulse repetition intervals) to solve the velocity ambiguity.

The performance of the FLR has been extended to very short distances and it has been tested in Stop&Go mode. The present design of the RF module has too much gain and the receiver saturates when the target is too close causing sometimes the loss of the target at close distances (<2m) which is not acceptable. In the new RF architecture under development, we plan not to use the LNA’s and also implement two levels of transmit power, so that the receivers will not saturate at close distances.

We had performed a wealth of data collection and testing with open loop test vehicles both in Israel (Toyota Previa) and in Europe (E-class Mercedes station). The FLR was tested in closed loop vehicles for long periods of time in BMW, Visteon and Siemens. We have also extensively integrated and tested the FLR in close collaboration with TNO in an Audi S8 test vehicle. The testing w
ith TNO allowed us to refine the Stop&Go algorithms.