# Formulation of a generalised switching CFAR with application to X-band maritime surveillance radar

- Graham V. Weinberg
^{1}Email author

**Received: **20 July 2015

**Accepted: **16 September 2015

**Published: **5 October 2015

## Abstract

A generalisation of a switching based detector is examined, allowing the construction of such detectors for target detection in any clutter model of interest. Such detectors are important in radar signal processing because they are robust solutions to the management of interference. Although formulated in general terms, the theory is applied to the design of a switching constant false alarm rate detector for X-band maritime surveillance radar. It is shown that such a detector manages the problem of interference better than standard detection processes.

## Keywords

## Background

Although the theory of CFAR detectors is well developed for the Exponentially distributed clutter case, there has been a requirement to re-examine the theory of CFAR as radar resolution has improved with corresponding advances in radar hardware design and processing throughput, resulting in improvements in radar resolution. Finer radar resolution is at the expense of spikier clutter returns, and an inevitable deviation from the validity of the Exponentially distributed clutter assumption (Shor and Levanon 1991; Anastassopoulos and Lampropoulos 1995). Consequently there has been much work on designing CFAR processes for more appropriate clutter models (Pourmottaghi et al. 2012; Weinberg 2013).

In the context of airborne X-band maritime surveillance radar, this is certainly the case, and recent work has established the Pareto clutter model assumption as valid (Farshchian and Posner 2010; Weinberg 2011). Thus there has been much research devoted to the design of CFAR detectors under a Pareto clutter model assumption (Weinberg 2013, 2014b).

Switching detectors were first examined in Cao (2008), which developed a switching CFAR based upon the cell-averaging (CA)-CFAR for target detection in Exponentially distributed clutter. It is shown in Cao (2008) that the performance of the CA-CFAR can be improved in the presence of interference through a switching process. This approach has thus generated much interest and stimulated ongoing research into the technique’s refinement and application to other clutter environments (Meng 2009; Erfanian and Vakili 2009; Tablet and Soltani 2009; Zhang et al. 2013).

The purpose of this paper is to show how the application of a switching CFAR to Pareto clutter introduced in Weinberg (2014b) can be used to formulate a general switching detector for application to any clutter model of interest. The key feature of a switching detector is that it pre-processes the clutter range profile to censor anomalous measurements of clutter. The approach in Weinberg (2014b) shows that the statistical structure of this censored clutter range profile, for the Pareto case, is equivalent to that as formulated in Cao (2008), for the Exponentially distributed clutter case. It will be shown that this can be generalised further, so that the censored set can be specified so that the properties of it for the Pareto case can be applied to an arbitrary clutter model.

Based upon this, a generalised switching detector is formulated, and a general expression for its probability of false alarm is derived, which can then be used to set threshold multipliers in practice. To illustrate this, a new switching detector is examined, based upon a Lomax distribution, which has become of much interest to researchers at Defence Science and Technology Group (DSTG) as a potential alternative to the standard Pareto model.

The paper is organised as follows “CFAR and transformed detectors” outlines CFAR from a mathematical perspective, and also overviews an appproach to the construction of CFARs based upon transformations. “Switching detector formulation” formulates the general switching CFAR, while “Example of switched detector” proposes a new switching CFAR based upon the Lomax clutter model assumption. “Ingara data and distributional fit” discusses the data set under consideration, including clutter model fitting. Finally “Application to X-band radar” provides some examples of detector’s performance, under a Lomax clutter model assumption, relative to non-switched detectors.

## CFAR and transformed detectors

*X*, which is also assumed to be independent of each member of the clutter range profile. A general detection scheme can be specified by

*g*computes a measure of the average level of clutter, and \(\tau\) is a threshold factor, which is used to regulate the false alarm probability. The notation used in (1) means that the null hypothesis is rejected if and only if \(X > g(X_1, X_2, \ldots , X_N; \tau ).\) The probability of false alarm (Pfa) of (1) is given by

These detectors can be transformed to operate in a clutter environment of interest using the transformation approach of Weinberg (2014), which generalised the Pareto case in Weinberg (2013). The elegant feature of this method is the detector is modified while the threshold multiplier (\(\sigma\) or \(\nu\)) is still set via the original Pfa expression. This method will be used to modify (3) and (4) to operate in Lomax distributed clutter in “Example of switched detector”; here we outline the general transformation approach for completeness.

*Y*, with distribution function \(F_Y(t):=\mathrm{{I\ P}}(Y\le t).\) In order to adapt the detection process (1) to operate in clutter modelled by random variables \(Y_1, Y_2, \ldots , Y_N\) with distribution function \(F_Y,\) one introduces a transfer function

## Switching detector formulation

Although this paper is focused on switching detectors, it is worth commenting on the fact that these are not the only solution to managing irregularities in clutter for CFAR processes. In particular, Smith and Varshney (1997) introduced a variability index (VI)-CFAR algorithm. This detector, designed to operate in independent Exponentially distributed clutter returns, has the CFAR property and is shown to manage interfering targets well. Additionally it is shown that it regulates the Pfa reasonably well. This detector partitions the clutter range profile into two sets, and runs a test to see whether there is variability in the two. It then tests to see whether the means of the two differ. Based upon these tests, and appropriate variation of the CA-CFAR is selected. An investigation revealed that when applied to the Pareto case, this detector becomes dependent on knowledge of the Pareto shape parameter, and hence is not a CFAR detector for the case of Pareto distributed clutter returns. Furthermore, even with the application of an maximum likelihood estimator for the Pareto shape parameter, the algorithm becomes computationally expensive to run in comparison to a switching detector.

The idea behind a switching detector is to determine whether any of the statistics in the clutter range profile can be classified as being irregular. Such irregularities in measurements could be due to clutter power level increases or spurious interfering targets. These irregularities, once identified, are censored from the detection process. Suppose \(S_0\) is a subset of the clutter range profile, consisting of clutter returns deemed not irregular. It will be shown how to select \(S_0\) so that the statistical properties of it in the Pareto case can be applied to arbitrary clutter models.

Let \(g_{|S_0}\) be the restriction of the clutter measurement function *g* to the set \(S_0.\) To illustrate this, suppose *g* is a sum, with clutter range profile \(\{Y_1, Y_2, \ldots , Y_N\}\) and that \(S_0 = \{Y_1, Y_3, Y_5\}.\) Then \(g_{|S_0}(Y_1, Y_2, \ldots , Y_N) = Y_1 + Y_3 + Y_5.\) If instead, *g* was an order statistic, then \(g_{|S_0}(Y_1, Y_2, \ldots , Y_N)\) would be the corresponding order statistic on the set \(\{Y_1, Y_3, Y_5\}.\)

It is clear that it is vital to determine a useful way in which to specify the set \(S_0.\) Once this is done, it is then possible to determine \(\kappa\) through an expression for the Pfa of (8).

*Z*is the CUT and \(a>0\) is a fixed constant Cao (2008). Selection of parameter

*a*is described comprehensively in Cao (2008) and Weinberg (2014b). This set can be used as a basis for all switching detectors with a simple application of the inverse of (6). The key to this is to reformulate the set \(S_0\) so that it is in terms of clutter statistics for the desired clutter model. To see this, it is shown in Weinberg (2014) that if

*Z*is an Exponentially distributed random variable with unity mean then

*H*, defined in (6), has the property that \(H(Z) \mathop {=}\limits ^{d}Y\) and \(Z \mathop {=}\limits ^{d}H^{-1}(Y),\) where

*Y*is the desired clutter model. Hence an application of these to the Exponentially distributed clutter range profile, and under the assumption of \(H_0,\)

*j*. Thus, (9) suggests an appropriate way in which clutter statistics may be sorted for a switching detector, based upon the clutter model of interest. Observe that if the set \(S_0\) is determined through (9), then any properties of \(S_0\) under \(H_0\) that have been established can be applied directly to the clutter environment of interest. In particular, it is shown in Weinberg (2014b) that the distribution of \(n_0\) under \(H_0\) is given by

*n*clutter statistics. Then, since clutter is sorted via (9), it follows by conditional probability that

*a*, used in the definition of \(S_0,\) can also be selected based upon the guidelines established in Weinberg (2014b).

## Example of switched detector

*h*(

*t*) has the properties that \(h(0) = 0\) and it is increasing, so that (12) is a well-defined distribution function. To illustrate this, the choice of \(h(t) = t\) produces the Exponential distribution, \(h(t) = t^n\) (with \(n>0\)) results in the Weibull family and \(h(t) = \log (t/\beta )\) yields a Pareto model. By applying (12) to (6), it can be shown that \(H(t) = h^{-1}(t/\mu )\) and consequently \(H^{-1}(t) = \mu h(t).\) Thus by applying these to (9), it follows for the model (12) that

*h*, which is applicable for any clutter model based upon (12).

*Y*has a Pareto distribution with shape parameter \(\alpha\) and scale parameter \(\beta.\) Then its distribution function is

Before examining the performance of the detector, a brief overview of the DSTG Ingara data is considered, together with an analysis of the Pareto and Lomax fits to this data.

## Ingara data and distributional fit

DSTG’s Ingara radar has provided researchers with X-band high resolution medium to high grazing angle polarimetric clutter returns that have assisted in the analysis of radar detection schemes (Stacy et al. 1996). This radar has been deployed in a series of trials, and subsequent analysis of the resultant clutter and detection performance has been documented extensively (Stacy et al. 2005; Crisp et al. 2007; Rosenberg 2010).

The clutter set used for the following analysis consists of a series of pure clutter returns obtained through a trial conducted by DSTG in 2004, and is designated run 34,683. The Ingara radar operated at X-Band, with a centre frequency of 10.1 GHz, and uncompressed pulse width of 20 μs, and range resolution of 0.75 m. The trial was run in the Southern Ocean, roughly 100 km south of Port Lincoln in South Australia. Further details of the Ingara radar can be found in Stacy et al. (1996), while a summary of the trial, and analysis of the resultant radar clutter, can be found in Stacy et al. (2005), Rosenberg (2010). The data was obtained through a number of runs, where in each such run the full 360° of azimuth angles were scanned. Thus the radar operated in a circular spotlight mode.

Analysis of Ingara data, and in particular the Pareto model fit, can be found in Weinberg (2011). Run 34,683 consists of 840,704 clutter returns, from the data set in the approximate up wind direction. This data set is thus the most spiky, and hence is the worst case scenario from a detection perspective. Only horizontal transmit and receive, and vertical transmit and receive, polarisations are considered. This is due to the fact that the vertically polarised case corresponds to approximate Rayleigh clutter amplitude statistics, while the horizontal polarised case consists of spikier clutter returns that are non-Rayleigh distributed (Weinberg 2011). Cross polarisation clutter tends to be distributed between these two extremes, and so the two polarisations considered represent the two extreme cases of clutter.

Statistical modes were fitted to these data sets via maximum likelihood estimation algorithms. In the fitting of the Pareto distribution to the horizontally polarised data set yielded estimates of \(\alpha = 4.7241\) and \(\beta = 0.0446,\) while in the vertically polarised case, these were \(\alpha = 11.3930\) and \(\beta = 0.3440.\) The Lomax distribution resulted in a shape parameter of \(\mu =84.8173\) for horizontal polarisation, and \(\mu = 31.2739\) in vertical polarisation.

The next section examines the performance of (18) relative to (16) and (17).

## Application to X-band radar

It is of interest to investigate whether the switching detector (18) improves on the performance of (16) when subjected to interference. Additionally, it is of interest to assess the performance of (18) relative to (17), since OS-based CFARs are robust to interference (Gandhi and Kassam 1988). This is done by simulating clutter from the model (15), where \(\mu\) is matched to estimates obtained from real radar clutter.

For the detector analysis, \(N=32\) with the Pfa set to 10^{−4}. The SW-CFAR uses \(a=1.5\) and \(N_{T} = 29,\) while the OS-CFAR uses \(k=30.\) The latter two choices are selected so that the respective detector can manage up to two interfering targets. Examining performance in both these polarisations enables guidelines on detection strategies to be formulated. Detection performance is estimated in all cases using \(10^6\) Monte Carlo samples for each signal to clutter ratio (SCR).

^{−4}, we see that in the case of no interference, the switching detector tends to improve on the detector (16) in both polarisations. In the horizontally polarised case, the switching detector has the largest deviation from the design Pfa. However, in the vertically polarised case, the switching detector is best at maintaining the desired Pfa.

Pfa estimates in uniform clutter, including when detectors are subjected to interference, through the insertion of a 5 dB Swerling 1 target into the clutter range profile

HH | VV | |||
---|---|---|---|---|

No inter | Inter | No inter | Inter | |

CA-CFAR | 0.94 × 10 | 5.5 × 10 | 1.07 × 10 | 5.5 × 10 |

OS-CFAR | 1.06 × 10 | 4.2 × 10 | 1.15 × 10 | 4 × 10 |

SW | 0.91 × 10 | 5.1 × 10 | 1.01 × 10 | 5.3 × 10 |

In the case of interference, the detector (17) performs the best in maintaining the desired Pfa. The switching detector only slightly improves on the performance on the detector (16). It is interesting to note that in the presence of interference, the resultant Pfa is smaller than the design Pfa. From a practical perspective, this is better than having an increase in the false alarm rate.

The results shown in Figs. 4, 5 and 6 demonstrate that the switching detector has a greater capacity to manage interfering targets than the non-switched detector on which it is based. It also has slightly better performance than an OS based detector. These results are common to the two polarisations. In terms of false alarm regulation, the switching detector tends to reduce the number of false alarms, especially in the presence of interference, as Fig. 6 shows.

## Conclusions

A general formulation of a switching detector was presented, and it was shown how the set \(S_0\) could be selected so that it became equivalent to that used for the case of Exponentially distributed clutter. This meant that setting the threshold multiplier for a switching detector could be done in a uniform way regardless of the underlying clutter model. Consequently, this generalised the Pareto case in Weinberg (2014b).

Due to issues with the Pareto clutter model, a Lomax distribution was examined, and a switched version of a transformed CA-CFAR was analysed. It was shown to provide a robust solution to the management of interference. Furthermore, its performance exceeded that of the non-switched equivalent, as well as that of an OS-CFAR. Further work will examine the application of such a detection scheme directly to real data.

## Declarations

### Compliance with ethical guidelines

**Competing interests** The author declares that he has no competing interests.

**Open Access**This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

## Authors’ Affiliations

## References

- Anastassopoulos V, Lampropoulos GA (1995) Optimal CFAR detection in Weibull clutter. IEEE Trans Aerospace Electron Syst 31:52–644View ArticleGoogle Scholar
- Blake S (1988) OS-CFAR theory for multiple targets and nonuniform clutter. IEEE Trans Aerospace Electron Syst 24:785–790View ArticleGoogle Scholar
- Cao T-TV (2008) Constant false alarm rate algorithm based upon test cell information. IET Radar Sonar Navig 2:200–213View ArticleGoogle Scholar
- Crisp DJ, Stacy NJS, Hudson DA, Pincus PB, Goh AS (2007) Polarimetric analysis of maritime SAR data collected with the DSTO Ingara X-band radar. In: Proceedings of the International Geoscience and Remote Sensing Symposium, pp 3870–3873Google Scholar
- Erfanian S, Vakili VT (2009) Introducing excision switching-CFAR in K-distributed sea clutter. Signal Process 89:1023–1031View ArticleGoogle Scholar
- Farshchian M, Posner FL (2010) The Pareto distribution for low grazing angle and high resolution X-band sea clutter. In: IEEE Radar Conference, pp 789–793Google Scholar
- Gandhi PP, Kassam SA (1988) Analysis of CFAR processors in nonhomogeneous background. IEEE Trans Aerospace Electron Syst 24:427–445View ArticleGoogle Scholar
- Goldstein GB (1973) False-alarm regulation in log-normal and Weibull clutter. IEEE Trans Aerospace Electron Syst (AES-9) 1:84–92View ArticleGoogle Scholar
- Meng XW (2009) Comments on constant false alarm rate algorithm based upon test cell information. IET Radar Sonar Navig 3:646–649View ArticleGoogle Scholar
- Minkler G, Minkler J (1990) CFAR: the principles of automatic radar detection in clutter. Magellan, BaltimoreGoogle Scholar
- Nagle DT, Saniie J (1995) Performance analysis of linearly combined order statistic CFAR detectors. IEEE Trans Aerospace Electron Syst 31:522–533View ArticleGoogle Scholar
- Nitzberg R (1979) Low-loss almost constant false-alarm rate processors. IEEE Trans Aerospace Electron Syst (AES-15) 5:719–723View ArticleGoogle Scholar
- Pourmottaghi A, Taban MR, Gazor S (2012) A CFAR detector in a nonhomogenous Weibull clutter. IEEE Trans Aerospace Electron Syst 48:1747–1758View ArticleGoogle Scholar
- Qin X, Zhou S, Zou H, Gao G (2013) A CFAR detection algorithm for generalized gamma distributed background in high-resolution SAR images. IEEE Geosci Remote Sens Lett 10:806–810View ArticleGoogle Scholar
- Rosenberg L, Crisp DJ, Stacy NJ (2010) Analysis of the KK-distribution with medium grazing angle sea-clutter. IET Radar Sonar Navig 4:209–222View ArticleGoogle Scholar
- Shor M, Levanon N (1991) Performances of order statistics CFAR. IEEE Trans Aerospace Electron Syst 27:214–224View ArticleGoogle Scholar
- Smith ME, Varshney PK (1997) VI-CFAR: a novel CFAR algorithm based on data variability. In: IEEE National Radar Conference, Syracuse, pp 263–268Google Scholar
- Stacy NJS, Burgess MP, Muller MR, Smith R (1996) Ingara: an integrated airborne imaging radar system. In: Proceedings of the International Geoscience and Remote Sensing Symposium, pp 1618–1620Google Scholar
- Stacy N, Crisp D, Goh A, Badger D, Preiss M (2005) Polarimetric analysis of fine resolution X-band sea clutter data. In: Proceedings of the International Geoscience and Remote Sensing Symposium, pp 2787–2790Google Scholar
- Tablet L, Soltani F (2009) A generalised switching CFAR processor based upon test cell statistics. Signal Image Video Process 3:265–273View ArticleGoogle Scholar
- Weinberg GV (2011) Assessing the Pareto fit to high resolution high grazing angle sea clutter. IET Electron Lett 47:516–517View ArticleGoogle Scholar
- Weinberg GV (2013) Constant false alarm rate detectors for Pareto clutter models. IET Radar Sonar Navig 7:153–163View ArticleGoogle Scholar
- Weinberg GV (2014) General transformation approach for constant false alarm rate detector development. Digital Signal Process 30:15–26View ArticleGoogle Scholar
- Weinberg GV (2014) Management of interference in Pareto CFAR processes using adaptive test cell analysis. Signal Process 104:264–273View ArticleGoogle Scholar
- Zhang R, Sheng W, Ma X (2013) Improved switching CFAR detector for non-homogeneous environments. Signal Process 93:35–48View ArticleGoogle Scholar
- Zhang R-I, Sheng W-X, Ma X-F, Han Y-B (2015) Clutter map CFAR detector based on maximal resolution cell. Signal Image Video Process 9:1151–1162View ArticleGoogle Scholar