1. Introduction

The Deployable Multiband Passive Active Radar (DMPAR) concept has been originally proposed by1,3-4 and the SET-152 RTG in the Final Report of the group6. In this approach, a collocated system comprising of four passive and active components has been connected through the algorithms of centralized and decentralized fusion of signals or plots, respectively. Performed simulations gave promising results and through the course of work of SET-1952 these were confirmed by experimental data. The scenarios from trials on which we registered signals and data are described in.2,11  The algorithms of signals and data fusion were developed on the basis described in.8-9

In this paper we use and shortly show centralized (fusion of signal statistics) and decentralized (fusion of detections) fusion strategy, perform analysis of system probability of false alarm and threshold management for centralized and decentralized processing. We recall system modelling and simulation which was presented in,8,10 present and discuss simulation results of detection for collocated passive-active system scenario10 and simulation results of the target location accuracy, comparing location and tracking accuracy, when (a) only passive radar is tracking the target, (b) both passive radar and the fire control radar are tracking the target.12

Based on results on registered signals and data described in2,9 we show results of detection for collocated and dislocated passive-active system scenario.

2. System processing scheme concept

Figure 1

Figure 1 Schematic depiction of analyzed system.

In this paper we analyze a system, that consists of high power, low frequency, long range radar (denoted as Early Warning Radar–EWR) and a set of collocated active and passive components–DMPAR (see Figure 1).

The scheme of data processing in the system has been presented on Figure 2. The first element is the EWR unit, which continuously scans the area for targets. It works independently with its own processing, detection and tracking algorithms. It is not perceived by the target as a threat, because its performance is low and does not satisfy requirements to engage the target successfully.

Our aim is to show performance of the passive components that are cued by EWR and relate it to performance of the active component working independently or in cooperation with passive elements. Although passive and active components are collocated, we assume that they are not necessarily mounted on one platform as was the case in,4,6 but they share the common surveillance area. We also assumed that active and passive components can be dislocated.2,11 What is more, we assume that we use our active component at a last resort or in the last moment of target engagement.

The scheme of data processing in the system has been presented on Figure 2. The first element is the EWR unit, which continuously scans the area for targets. It works independently with its own processing, detection and tracking algorithms. It is not perceived by the target as a threat, because its performance is low and does not satisfy requirements to engage the target successfully.

Figure 2

Figure 2 Overview of data processing in the system.

The EWR outputs its data in a form of tracks and sends it to the collocated DMPAR system (by means of Command and Control system).

Our described system may work in three modes of operation, specifically:

  • in mode A, only the passive radars are operational and are being cued by the EWR. The active component is activated during the last second before the engagement
  • in mode B, only active radar is operational and cued by the EWR. This mode is not preferable because it alarms the target that it may be engaged
  • in mode C both active and passive components are operational, although this mode is also not preferable due to the same reason as mode B.

Due to the operational requirements, modes B and C may not be used at all, as they can warn the target of the presence of high-fidelity active component (i.e. a Fire Control Radar) and it may break its planned course of action and steer out of the engagement zone.

Let us then start with mode A processing scheme. As cueing information is fed into each channel of the passive component (each transmitter-receiver pair) the data from cued regions is provided on a per-target basis to centralized detection module (see Figure 3). In the same time each sensor continues its operation and all sensor detections are sent to the decentralized detection module. This way of operation allows for simultaneous operation of data and signal fusion on separable areas of detection grid.

Figure 3

Figure 3 Simultaneous operation of both centralized and decentralized fusion.

Thanks to the feeding the passive radar by the exogenous radar tracks (cueing by EWR), covert tracking of aerial targets can be continued by the centralized fusion excluding the decentralized fusion. The added value of such approach is the extended range of passive radar that is not limited to one of its working channels. If the target is located within the operational range of the passive radar and has not been detected by the EWR, the decentralized fusion is performed as well and may still provide detection and tracking of the target.

2.1. Centralized and decentralized fusion strategy

Following the work presented in4,6 we can divide our approaches to data fusion to two algorithms:

  • decentralized fusion on the level of plots (detections)
  • centralized fusion on the level of signals (signal statistics).

In case of the decentralized fusion each of the sensors in the sensor network operates autonomously. The detection process is performed on the sensor, utilizing any given algorithm that leads to a binary decision Dd∈{0,1} where 0 indicates no detection of target and 1 indicates a positive detection. Probability of false alarm Pfa is the only constraint imposed by the sensor network. At the central decision node, detections from each sensor are collected and central decision is made following the rule k out of m, where  is the total number of participating sensors and k is minimum number to indicate a positive detection.

In the centralized fusion scheme, at the central decision node a combined statistic of signal in each channel is calculated:

Equation 1      (1)

where weighting factor Inline 1, Inline 2 is the SNR at the tested position, N is the overall mean noise floor level in the vicinity of the tested position and di is the amplitude of the signal at the cell under test. This statistic is then compared to a predefined threshold, which guarantees certain probability of false alarm.

2.3. Analysis of system probability of false alarm

The main idea behind the proposed system is to achieve the largest probability of detection while maintaining the desired probability of false track confirmation.

Let us assume two disjoint areas in the passive radar search volume:

  • "cued volume" understood as the targets and its’ vicinities cued by EWR where centralized fusion takes place
  • "search volume" as the complement to the cued area where only decentralized fusion takes place

Equation 2      (2)


Inline 3 – probability of false target detection in single refresh interval,

Inline 4 – probability of false target detection in search volume region,

Inline 5 – probability of false target detection in cued volume region,

ρ – capture area to searching volume ratio.

If the cueing volume is much smaller than the searching volume (as usually is the case) we can gain the additional benefit by increasing the detection probability (by lowering the threshold) without significant effect on increasing the probability of false track confirmation:

Equation 3      (3)

Concluding: after capturing the target from the EWR, the rule "k out of m" in the consecutive refresh periods is performed to confirm the track of the cued target.

As one can easily see, if ρ is small, the Inline 5, even if increased significantly, has a small contribution to the probability of false track confirmation. It allows to decrease the threshold level in the cued regions and thereby increase the probability of detection of the target.

2.2. Threshold management for centralized and decentralized processing

To calculate Inline 3 given a target value of  Inline 7–probability of false track confirmation, we can utilize an analysis performed for SET-195 report:2

Equation 4(4)


Inline 7 – probability of false track confirmation

Inline 8 – number of consecutive periods to wait for track confirmation

Inline 9 – number of required detection to confirm a track.

Threshold for both the centralized and decentralized processing scheme is set according to calculated Inline 3. For the centralized fusion case we would need to know the distribution of statistic D in case there is no echo in the signal. This can be estimated with gathering data from regions free of objects and processing them with the centralized algorithm.

The decentralized processing scheme forces each sensor to perform detection with a given–higher than  Inline 3–probability of false alarm, namely Inline 10, which can be calculated as:

Equation 5      (5)


N – total number of sensor channels in the system

K – number of positive detections of the target among channels

assuming same false alarm probabilities between channels. It should be noted that keeping same false alarm probabilities between channels has its drawback, as it is suboptimal taking probabilities of detection in each channel into consideration.3,8

3. System modelling and simulation

3.1. Passive radar model

Implemented simulation consists of signal generation in the Illuminator of Opportunity (IO), it is propagation and all main steps of passive radar receiver--digital beamforming, direct signal cancellation and correlation processing.7 Passive radar receiver is capable of processing FM and DVB-T signals. The first step in simulation is the generation of signal in FM and DVB-T transmitters with suitable signal characteristics. Subsequently the direct signal is captured by reference channel and the echo signal is captured by surveillance channel of the passive DMPAR subsystem. The direct signal is removed from the surveillance channel by adaptive filter. The correlation process and threshold management enables target detection, centralized and decentralized fusion. Although centralized algorithm may be clearly implementable due to existing cueing, the decentralized model implementation needs a data correlation algorithm to properly assign detections from different sensors with each other. Such an algorithm is out of scope of this article, however for the simulations we have implemented a model which gives us an upper bound performance of decentralized fusion by always making the right association.

3.2. Simulated scenario and measured statistics

Figure 4

Figure 4 The situation view of the simulated scenario.

DMPAR system is mounted in the middle of coordinate system. The target of interest is flying at the altitude of 2500 m in the west direction with the velocity of 60m/s. The IOs are depicted in the Figure 4. The exact IOs utilization is scenario dependent. Results are being compared to an active Fire Control Radar component, being part of the sensor network.

In scenario 1, we utilize DVB-T IOs one to four. We can predict that comparable bistatic range of each transmitter–receiver pair allows us to show increased coverage, due to fusion processing.

In scenario 2, we utilize additional FM transmitter. The detection range of the system is greatly extended, since the FM channel has theoretical range lying about 450 km from the DMPAR system for targets at high altitudes (in case of low flying objects radar horizon range is shortened). In this scenario we can estimate how much one dominant sensor can extend the range of the system.

As a result of the simulations the probability of detection/localization/track confirmation statistic is calculated, depending on which sensor and which type of processing is used vs range. The resulting range is converted to monostatic equivalent along the target path.

Figure 5

Figure 5 Comparison of probabilities of detection for 4 channels of DVB-T sensors and one FM channel. Horizon shadowing has been accounted for in the simulation.

3.3. Simulation results in detection

On the Figures 5 to 7 combined results from simulations are presented. First, a common result from scenario 1 and 2 in terms of probability of detection for sensors 1 to 5 is presented on Figure 5. As one can see 4 DVB-T sensors have comparable ranges at about 75 to 85 km at the level of Pd = 0.9. The added FM sensor has much greater range, but limited by horizontal shadowing of the target at around 200 km.

Figure 6

Figure 6 Comparison of probabilities of localization/track confirmation for 4 channels of DVB-T sensors. Horizon shadowing has been accounted for in the simulation.

According to scenario 1 the detection and localization is made by the EWR and 3D track is passed to PCL networked sensor working, as described in 3.1 in one of three modes. As one can notice, the standard PCL mode of operation, requiring autonomous detection by at least 3 PCL Tx-Rx pairs gives the worst result in terms of range. What is clearly visible is that cued target can be detected by the system with DMPAR processing at much longer range than in standard mode of operation. The gain in range is evident as we move from 85 km without fusion through 112,5 km with decentralized and 125 km with centralized processing.

Figure 7

Figure 7 Comparison of probabilities of localization/track confirmation for 4 channels of DVB-T sensors and one FM channel. Horizon shadowing has been accounted for in the simulation.

The second scenario presents a special case of processing with a single dominant channel and a set of channels with much lower range. It is evident that in case of centralized processing the aforementioned channel draws up the statistic D, regardless of other sensors range. In case of decentralized processing a single dominant sensor is not enough to significantly boost the detection range. That is because each of the detection in decentralized scheme is given the same weight regardless of  SNR and here the worse DVB-T channels start to take the dominant position thus limiting the effective range. The same effect has been observed during processing data from DETOUR trials from SET-195 on "DMPAR Short Term Solution Verification" as can be read in.2,11

3.4. Simulation results in location

Another look at the cued PCL performance is from the location and tracking accuracy perspective. To compare these statistics we have analyzed the standard, SNR-based equations for range and doppler estimation performance.13 Next, we have simulated PCL cartesian localization process and measured resulting errors. Our results, expressed as x-y and z–coordinate location error are presented on Figures 8a and b.

Figure 8

Figure 8 a (left) and b (right) Cartesian location performance results in terms of a) sigma f x-y error; b) sigma of z error.

These results were achieved utilizing four DVB-T Rx-Tx pairs. As can easily be predicted, fusion algorithm does not give a direct gain in location performance, because it does not provide direct SNR gain. The gain from applying cued PCL approach becomes visible when analyzing tracking performance, as earlier initiation of tracking provides more time to acquire target and prepare best reaction. Example results are presented on figure x a and b.

Figure 9

Figure 9 a (left) and b compared to the active FCR radar (a – left) and cartesian tracking performance in terms of z coordinate compared between FCR and cued PCL (b – right).

On Figure 9a, performance of cued PCL is compared to the FCR in terms of elevation error. As we can see, cued passive component underperforms compared to the active device. However, as we can note on the Figure 9b, earlier start of tracking due to the centralized processing can help outperform the active radar.

4. Chosen results of DETOUR trials

Figure 10

Figure 10 Flight FL2 trajectory. For analysis was chosen part of the flight between points A and G.

In the activities of the NATO task group SET 195 "DMPAR Short Term Solution Verification" on the Czech airport near Hradec Kralove field trials were conducted in order to collect data for further analysis on the main group purpose. At the beginning of September 2014 during 4-days field trials five institutions, both domestic and foreign, took part in research.  

In order to perform multiband fusion algorithms a subset of the whole raw data registered by each of the sensors was chosen and processed.  Detections from two flight trajectories registered by all the sensors were chosen to perform fusion algorithms on real data. On image (see Figure 10) maps with flight trajectories (FL2) and positions of radar sensors are shown.

Figure 11

Figure 11 Comparison of detection range in collocated configuration, flight FL2.

On the Figure 11 comparison of centralized and decentralized processing in case of collocated antennas (active S-, C-Band and passive FM, DVBT) is presented. One can note the lower performance of the decentralized processing comparing to both centralized and S-Band only processing. This is due to significant number of sensors with low SNR (nine passive channels) shadowing relatively small number of sensors with high SNR (two active channels). Due to the k out of m rule of decentralized detection, active radar performance is lost in the number of passive channels. With the centralized processing, low-SNR channels are scaled down in the threshold statistic which gives way for the signals from active component with high SNR. In case of far-range detections, most of the channels start to face low SNR levels, and decentralized processing outperforms centralized scheme.

Figure 12

Figure 12 Comparison of detection range in dislocated configuration, flight FL2.

On the Figure 12 the comparison of centralized and decentralized processing in case of dislocated antennas (all channels in Tab. 2) is presented. In contrary to the collocated case, decentralized processing outperforms centralized in most part of the range. This may be due to the fact, that far-range region for the active, ERA-FM and FFI-DVB-T components is close-range to the sensors in Jaromer or Chotec. Then, at least half of the sensors have reasonable SNR levels, allowing for decentralized scheme to work, whereas SNRs are still low for the centralized scheme to work (far-range and shadowing for active component especially). The coverage in the far-range is mostly provided by WUT-FM sensor based in Jaromer. In the dislocated case, the detection performance is influenced by target type and its trajectory as well as system configuration and receivers and transmitters positions.

Table 1  List of Track Visibility (TV) parameter for all cases in flight FL2


Flight FL2
Analysed case

Decentralized processing

Centralized processing


real TV


real TV














In the same time, Track Visibility (TV) is about 10 percent higher for centralized processing (Tab. 3). In dislocated sensors scenario, the comparison of decentralized and centralized processing performance  leads to a conclusion that detection range and TV are dependent on the type of track, specific terrain setting and PCL receiver type.

5. Summary

This paper demonstrates a novel approach to multistatic passive-active system like DMPAR system, which places the described concept in air and missile defense operational context. As SET-195 team has proven the applicability of "DMPAR Short Term Solution Verification," it makes implementation of presented processing in real system even more likely.

The presented radar models have been implemented and a number of simulations has been conducted. Elementary scenario consisted of a point target which track was transferred (cued) from EWR to passive radar. In each case the probability of detection as a function of target range has been evaluated. As we can note also on the Figure 9b, thanks to cueing, earlier start of tracking due to the centralized processing can help outperform the active radar. It is important to underline that, PCL systems is capable to track more cued objects without performance loss, because resource allocation is not needed.

The results from simulation and DETOUR trials are very promising as we can improve the performance of PCL sensor system to almost match this of active Fire Control Radar, while at the same time remaining covert and undetectable to potential threat. All in all this opens up a new line of research in passive-active air and missile defense systems.


  1. T. Brenner, J. Hardejewicz, M. Nałęcz and M. Rupniewski, "Signals and Data Fusion in a Two-Band Radar," Proc. Inter. Radar Symposium (IRS), Warsaw, 2012, pp.15-18.
  2. "SET-195 RTG Technical Report: DMPAR Short Term Solution Verification," NATO STO publishing, STO-TR-SET-195.
  3. M. Rupniewski, "Simulation Results of Decentralized and Centralized Detection of Targets in Two Band Radar," Institute of Electronic Systems, Warsaw, April 2011.
  4. T. Brenner, G. Weiss, M. Klein and H. Kuschel, "Signals and Data Fusion in a Deployable Multiband Passive-Active Radar (DMPAR)," Proc. IET Inter. Conf. on Radar Systems, Glasgow, 2012, pp.1–6.
  5. A. Farina and M. Grazzini, "Radar Fusion to Detect Dim Targets," Signal Processing 80(9), September 2000, pp.1833-1847.
  6. "SET-152 RTG Technical Report: Deployable Multi-Band Passive/Active Radar for Air Defense," NATO STO publishing, RTO-TR-SET-152, 2014.
  7. M. Malanowski, "Signal Processing Algorithms and Target Localization Methods for Continuous-Wave Passive Bistatic Radar," Oficyna Wydawnicza Politechniki Warszawskiej, Warsaw, 2012.
  8.  T. Brenner, L. Lamentowski and M. Rupniewski, "Processing of Measurements in DMPAR," 10th meeting of SET 195, Prague, November 2015.
  9. T. Brenner, L. Lamentowski , L. Raczyński and  L. Żochowski, "Off-Line DMPAR Registered Signal Processing and Analysis of the Results," RSM SET 231, Lisbone, October 2016.
  10. T.  Brenner, L. Lamentowski and R. Mularzuk, "Enhanced Target Detection by Cueing and Threshold Management in Sensor Network with DMPAR Processing," IRS 2017, Prague, June 2017.
  11. V. Stejskal, H. Kuschel, T. Brenner, L. Lamentowski, I. Norheim-Næss, P. Samczynski and J. Kulpa, "DETOUR Trials: The Mission and Its Results," IRS 2017, Prague, June 2017.
  12. T. Brenner, L. Lamentowski and R. Mularzuk, "The Analysis of Target Location Accuracy Enhancement via Cueing in Sensor Network with DMPAR Processing," PCL Focus Day, Fraunhofer FHR, May 2017.
  13. R. G. Curry, "Radar System Performance Modelling," 2nd Edition, Artech House, 2005.