DRS05 in Odometry and Speed & Distance Unit

The DRS05 radar sensors are based on a proven technology and meet the highest quality standards. The housing shape and the housing thickness withstand stone impact even at maximum speed. They are equipped with two microwave modules whose signals are combined by a patented algorithm. This improves the accuracy on changing track ground while the comparison of the signals evaluated in parallel contributes to the safety. If one of the modules fails or if the antenna receives a weaker reflected signal, the firmware changes to a special “single mode algorithm” to increase the availability of the speed signal.

An antenna module emits a fixed-frequency continuous wave (24 GHz, 5 mW) in a slanted beam towards the track ground, which ideally scatters the wave diffusely. The back-scattered wave whose frequency is shifted by the movement is multiplied by the transmitted signal to yield the Doppler signal of difference frequencies. Its spectrum contains a line whose position is proportional to the speed and whose width reflects the directional characteristic of the antenna – ideally. In fact, the track ground also influences the angle dependence of the backscattering; but only minimally with the DRS05 as said before. By the way, each antenna module provides a signal pair like a two-channel incremental pulse generator for determining the driving direction (DOT).

Typically, we mount two DRS05 units under the car body close to the bogie – but not on the bogie because the vibrations would shorten the service life of the sensor.

Typically, the odometer is an algorithm executed on the EVC (EVC = European Vital Computer, core of the ETCS on-board equipment). In a safe and highly available manner, it is to provide the speed and the distance travelled from multiple sensor signals which individually are insufficient. And because the signals of similar sensors are often disturbed simultaneously, safety can only be ensured using diverse sensors.

Typical diverse systems count on axle- mounted sensors in combination with our Doppler radar sensor DRS05 – a simultaneously occurring error of radar sensor and axle-mounted sensors is deemed unlikely. The DRS05’s precision being stable for decades allows to calibrate wheel diameters dynamically. It allows to increase the balise distances on high-speed routes and decrease train headway in metros. Although there are some ETCS projects at the moment where attempts are made to establish diversity by using axle-mounted sensors and an alternative sensor technology. The results are not at all satisfactory: Compared to optical sensors, radar in the K-band is insensitive to soiling; compared to satellite navigation, a radar sensor is highly available, even in street canyons and tunnels; and the signals of acceleration sensors cannot be integrated over a sufficiently long period of time due to the uncertainty of the inclination of the track (clue ’cross-sensitivity to acceleration of gravity’). Therefore, the high market penetration of the Doppler radar DRS05 continues to grow.

The motivation for developing the SDU library to be integrated in the customer software was to help our customers to maximise the benefits of our sensor products. Odometry is not an easy craft. The measurement uncertainty cannot be described adequately by white noise if the target is SIL4. There often is a narrow Gaussian part of the distribution which greatly underestimates the actual frequency in case of major deviations. In case of two grossly different measured values, a Gaussian fault model would provide a mean result which would be clearly different from the two measured values, although it is more likely that only one of the two measured values is grossly incorrect – or both in the same direction; e.g., in case of slip or slide. In addition, there are correlations in time, namely error states. Finally, diagnostic information is to be taken into consideration which is already provided to some extent by the sensor in case of DRS05. In general, however, diagnostic information shall be a by-product of the odometer to take advantage of the other sensors' signals. Some proposals of customers for extended functionalities of the DRS05 turned out to be a workaround for special situations which the odometry failed at.

In the railway sector, the state-of-theart technology used were Kalman filters; i.e., Bayes estimators for the train’s state of movement with Gaussian fault models for the sensors, in combination with various heuristics for the selection of the sensor signals. The state machine is complex to analyse, and with each addition to the fault model of a sensor, the entire algorithm needs to be reconsidered. Furthermore the width of the confidence interval to be issued is defined a priori depending on the remaining sensor configuration resulting from the application of the heuristics (keyword: degraded mode). Afterwards, when analysing the records, it is often found that the confidence interval was sometimes too small; i.e., uncertain, sometimes too large; i.e., bad for the performance.

I have replaced the heuristics by a Bayesian treatment of the outliers as well as the error states. By declaring the fault models of the sensors – one or two DRS05s and two at least two-channel axle-mounted sensors – to be an input of the algorithm, its correctness could be mathematically proven which greatly simplified the safety case. With a Bayes filter that can also represent multimodal probability distributions for the speed and acceleration of the train, we managed to calculate the confidence interval adapted to the situation which, as a result, usually is narrower than with a conventional odometry but safer.