Cognition and Radar Systems

C. J. Baker Ohio State University Columbus, Ohio, USA

chrisjbaker1@googlemail.com

Abstract — Cognition means ‘kowning’ and is the process by which we know about the world.. In human beings this is extremely complicated and although much is understood there is much more to be learnt. However, humans and perhaps more pertinently for radar systems, echo locating mammals are able to sense and interact with their environment in sophisticated ways that are beyond synthetic systems. In this paper we examine the nature, role and value of cognition in radar systems

I. INTRODUCTION

Cognition in mammals typically involves the interplay between multiple sensory receptors and neuronal operations of immense interrelated complexity and magnitude. It is likely this complexity allows such sophistication of action and interaction. One example that would be highly desirable to mimic with synthetic systems is autonomy which has proved challenging or radar based sensing systems to date. Radars sense their environment using electro-magnetic echo location. The information acquired may be thought of as uneven shaped voxels of reflectivity in angle, angle and radial range space usually updated on a regular or deterministic basis. This information might be plotted on a PPI display as in the case of air traffic control radars [e.g. 1] or as a high resolution map like image in the case of SAR [e.g. 2] and there are many other examples. Figure 1 shows an example of a primary air traffic control radar. This rotates in the azimuth plane detection and mapping out the position of approaching and receding aircraft. However, unless there is some means of dynamically and repeatedly making decisions on the basis of the information received the sensor serves no purpose. In simple radar systems, such as air traffic control decision making is carried out by an operator who continually monitors aircraft poison in 3-D space to ensure that they are able to take-off and land in safety. It is this latter function that turns the radar into a useful device. It is here the term radar system embracing the operator becomes dominantly significant. Thus we have, in air traffic control, a cognitive system only when there is a human operator. I.e. It is the human who perceives the skies and the position of aircraft, whilst the radar merely acts as the ‘sensory receptor’. Indeed cognition is supplied by the human operator at a level far beyond that of any synthesized for. However, the synthesis of equivalent

performance would enable a much wider set of applications to be adopted such as truly autonomous bevaior.

Further, consider the case of an electronically scanned array radar sensor. The advantages of electronic scanning are well known and can be crudely summarized as enabling the radar to ‘see’ anywhere at any time with an ability to change on a millisecond time scale. Indeed, they have more advanced properties than this and can, for example, even look in multiple directions simultaneously. Again, setting aside many details and subtleties, this can be thought of as the radar sensor having sensing abilities that a human operator is insufficiently quick to react to. This naturally implies a need to synthesize aspects of the human into the signal processing and control of the radar sensor. Ultimately, to achieve the full, latent, potential of electronic scanning the radar has to develop this area of cognition. Therefore the role of the human will still be one of an editorial nature, still interpreting the displayed information and passing instruction. Ultimately, if truly cognitive processing systems emerge then potentially the human operator could be removed altogether. However, the challenges of developing robust synthetic cognitive sensors of this sophistication are currently immense.

Figure 1. Example of an Air Traffic Control radar

1326

In the above we have used the example of electronic scanning but not explicitly considered other parameter variables. There are sensor parameters such as Power, beam width, PRF, pulse width waveform modulation, polarization etc. We can also exploit the relative and changing positions between the radar and the object or environment being sensed. That is to say the information acquired is a function of the acquisition geometry and time. A simple example is the detection of a target that is in the shadow of another object. A change in the sensor position may reveal the target to the radar and detection can again commence. This is a rarely used degree of freedom within the radar system (i.e. not the sensor) but is done with aplomb in the natural world of echo locating mammals [3].

There have been some early forays into this challenging but key area for future radar systems. Publications include general expositions such as [4] and more direct application as in [5, 6, 7]. In addition there are related research examples including knowledge based systems [8]. Here, the opportunity is taken to consider the nature of intelligence and reach out to some of the pioneering work emerging from disciplines such as artificial intelligence and cognitive signal processing. Given this context we examine some specific examples of how intelligence might be developed in a holistic manner taking cues from observations in the natural world. In so doing more questions than answers are raised but in a manner that suggests further development.

II. COGNITION AND INTELLIGENCE

In order to develop and incorporate intelligence into future radar design it makes logical sense to pose the question, ‘what do we mean by the term intelligence’? However, especially in the world of artificial intelligence, there doesn’t seem to be a universally accepted definition. The dictionary definition [9] says ‘the ability to understand, learn, and think things out quickly’. These are all good attributes we might wish to ascribe to a radar system but themselves pose more questions that they answer. More tellingly Gregory [10] states that’ Innumerable tests are available for measuring intelligence, yet no on is quite certain of what intelligence is, or even just what it is the available tests are measuring. In [11] Hutter attempts to define machine intelligence using a combination of a textural informal definition and a more formal mathematical description. The informal definition is stated as ‘Intelligence measures an agent’s ability to achieve goals in a wide variety of environments’. Whilst being self contained and consistent it casts intelligence as being something that has measurable properties that define it as with most phenomena. The formal description is cats mathematically, boldly attempting to capture an agents general ability to achieve goals in a wide range of environments. This is potentially very helpful but to make practical progress we need to consider both the dictionary and the more formal definitions to see that intelligent behavior has certain desired outcomes and that intelligence is a measure of how well they are achieved. Progress on both fronts is required. Indeed, Haykin [4] uses the term cognition with its implication of perception, intuition and reasoning and this maybe more helpful.

III. COGNITION AND RADAR SYSTEMS

We have already presented the example of an air traffic control radar system embracing cognition in the form of the radar operator. Indeed without the human interpreting the picture, communicating instructions to the pilots and observing their re-positioning, not only would the system exhibit no intelligence but it would be completely useless. Thus it must be concluded that cognition is not an option in a radar system it is, in fact, mandatory. The issue is how and in particular how much of the human operator can be replaced by cognition built into the signal processing. Again, this is bought into sharp focus for electronically scanned radar systems where the radar is able to act and re-act on a faster timescale than can a human being.

From the above we can usefully begin to divide radar system cognition into two differing forms within the radar system. The first is in the setting and re-setting of radar parameters on a pulse by pulse basis with the aim of maximizing information received against that desired. This presupposes that the radar has been ‘tasked’ in some form. This tasking of the radar and the role this has in defining the required degree of cognition is critical but is often absent. However, it is this tasking that provides the imperative against which the radar seeks to achieve success. Coming back to the air traffic control example the operator has clear metrics in terms of safety distances, landing rates etc. These translate to a tasking environment that is relatively well controlled. Indeed, even in the case of unexpected and unusual behavior there is a script that is followed with assurance of safety always paramount. This is in stark contrast to an air defense scenario where, implicitly, hostile air targets are attempting to avoid any adherence to a script and tasking is much less clear, dynamic and open ended.

Nevertheless, in both situations the creation of the ‘best’ information is something that parametric adjustment can help with. Figure 1 show a schematic rendition of an electronically scanned radar sensor on board a naval vessel faced with a somewhat contrived but demanding scenario.

Figure 2. Schematic description of targets to be detected and tarcked by electronically scanned radar

It is quite clear that the different tasks and the different distribution of targets calls for individually tailored radar

1327

parameters to maximize performance based in an assessment of prioritization. It is equally clear that the resources and resource timeline is finite and to optimize performance has to be carefully utilized. Consider the example of the need for update rates on a receding target near to the limit of the detection range. The updates will undoubtedly be a lot lower than one coming straight towards you at near range and high velocity!! A simple and perhaps zonally controlled setting of the PRF would ensure that less radar resources are used for the far target than the near. This can be taken further by introducing techniques such as fuzzy logic which can assign priority levels such that the PRFs set in both cases can be ‘softly’ altered in relation to the behavior and tasking of the system. Figure 2 shows fuzzy logic prioritization setting. We also note that now we are starting to build cognition into the radar system via the way in which the priorities in the fuzzy logic are being set. We would expect quite different settings for the air traffic control and air defense cases.

Time (sec)

Prio

rity

Target Priority

Figure 3. Example of fuzzy logic priority setting

It is probably accurate to say that a system which exhibits such control over the setting of its operating parameters in order to achieve a level of performance in this way is intelligent in a way that is consistent with the dictionary definition. However, if that were all that happened, as in the case of an operatorless air traffic control radar, it would still be of little value. The role of the operator is to interpret the observed environment and then effect a change in order to achieve a desired outcome. In the case of a naval vessel under attack this may be to launch a missile or deploy countermeasures with the intent of preserving own life. A successful missile launch would remove the hostile attack craft from the environment and the radar would redeploy its resources, again with the aim of achieving a desired picture of air activity. This second area of cognition that aims to bring about a desired effect exhibits a much more sophisticated and complex form of cognition and one that presents a much greater challenge.

Figure 4. Feedback loops for cognitive radar systems

Figure 3 attempts to illustrate this schematically building on the view of cognitive radar as a closed-loop dynamic system as expressed in [4]. Here we separate out the sensor parametric adjustment from the broader action and decision space that creates the desired change in the observed environment. It is likely that much greater progress can be usefully made in the inner ‘sensor’ loop and the articulation of best picture by synthetic cognition. The outer ‘action’ loop present a much more significant challenge but progress is being made in areas such as computer vision using optical imagery [e.g. 12].

IV. NATURALY INTLELLIGENT RADAR

There are numerous examples of echo location being used with great success in the natural world. Perhaps the most well known examples are those of the bat, whale and dolphin. Here we limit ourselves to the bat which exhibits many characteristics that would be desirable to replicate in systems using radar. Bats have also had over 50 million years of evolution to hone themselves as holistic sensing systems of extraordinary ability.

Figures 4 and 5 show two calls in sequence made by a bat during the course of feeding from an insect. The insect was balanced on a pin and hence is not ‘on the wing’ but a remarkable number of observations can be made regarding the parameter selection during detection and attack of prey.

Figure 5. Eptesicus nilsonii: search phase pulse analysis. The first chirp of the time series is analysed in its Time Domain representation, power

spectrum, spectrogram, WAF, range and Doppler profiles

1328

The spectrogram of the call (emitted pulse) made early in the engagement shows three harmonics each broadly with a longish constant frequency component but with a frequency modulation downswing in the early part of the pulse. The ambiguity function exhibits properties showing a resolution in both the range and Doppler domains. As the bat begins the process of intercepting the insect (on the top right you can see the entire sequence of pulses) a number of changes occur. Firstly there are two distinct regimes of PRF. One low,, in the early phase and one high, in the later phase. In this example the PRF is approximately constant but in others this is not the case. However, the flight time between pulses is normally long enough for the bat to ‘re-position’ it’s sensor via head and/or body turning and hence the sampling is far from uniform. Indeed the bat seems to be using information gleaned from predecessor pulses to take up a position of maximum advantage most likely a combination of detection, classification and location (but of course it’s hard to ask a bat).

Figure 5 shows an example pulse from the final high PRF phase of the ‘mission’. Again, if we start by examining the spectrogram we see that the form of the modulation has changed considerably. The pulse length has shortened greatly (why transmit lots of energy when you don’t need to) and the form of the frequency modulation is significantly different. It is close to hyperbolic with little or no constant component. This is mirrored by an alteration in the form of the ambiguity function whereby range resolution is high but there is little or no Doppler resolution. We might conclude therefore, that the bat has only need for ranging information (perhaps for aim point selection) and the momentum to intercept precludes the need for Doppler.

Figure 6. Eptesicus nilsonii: approach phase pulse analysis

These two pulse examples show all sorts of characteristics and use of parameter freedoms not normally utilized and exploited by radar systems. However, the technology does exist today to mimic such pulse and waveform agility. The question remains as to how? Of course part of the answer to this question sits between the two ears of the bats, I.e. in the cognitive processor. Whilst neurological research has and is being conducted there is far more that is not understood about the working of the bats brain than is understood. The study of such high performance naturally occurring systems offers a

rich environment from which many valuable lessons may be learnt and subsequently applied to take radar systems down a route towards the adoption of truly intelligent sensor systems.

V. CONCLUSIONS

Cognition in current radar sensors is, at best, at a very low level and often nonexistent. The cognition in radar systems is provided by a human being and represents the very best that is possible. In this paper we have specifically drawn a distinction between these two descriptions in order to emphasize differing aspects of cognition as it might apply to future sensors and senor systems. The dynamical parameter and ‘platform’ variation exploited by echo locating bats when foraging and feeding (as well as transiting in known and unknown locales) offers a learning environment for mimicking and embedding more intelligent behavior into radar systems. Often this will include platform relocation as well as variation in waveform selection as a function of space and time. To develop truly cognitive radar systems there remains much to be done but to realize the true potential of technology advances such as electronic scanning there is little option but to pursue this route.

REFERENCES

[1] S. Kingsley and S Quegan,, “Undertanding radar systems,” Sci Tech publishing, 1999.

[2] J. C. Curlander and R. N. McDonough, “Synthetic Aperture Radar”, John Wiley and sons, 1991.

[3] M. W. Holderoid, C. J. Baker, M. Vespe and G. Jones “Understanding signal design during the pursuit of aerial insects by echolocating bats: tools and applications,”Integrative and comparative biology, pp 1-11 May 2008.

[4] S. Haykin, “Cognitive radar,” in Knowledge based radar detection, tracking and classification, F. Gini and M. Rangaswamy, Eds. John Wiley and sons, 2008, pp.9-30.

[5] S. L. Miranda, C. J. Baker, K. Woddbridge and H. D Griffiths, “Mutlifunction radar resource management,” in Knowledge based radar detection, tracking and classification, F. Gini and M. Rangaswamy, Eds. John Wiley and sons, 2008, pp.225-264.

[6] F. Barberesco, P.O. Gaillard, T. Guenais, N. Harvey, M. Johnston, B. Monnier, P. Sims and R. Wills, “Intelligent radar management by advancing beam scheduling algorithms based on short time planning and contstraints relaxation strategies,” IEEE conference on Phased array systems and technology, pp 283-288, Oct 2003.

[7] F. Smits, A. Huizing, W. van Rossum and P. Hiemstra, “A cognitive radar network: Architecture and application to multiplatform radar management,” EuRad 2008, pp 312-315, Oct, 2008.

[8] F. Gini and M. Rangaswamy, “Knowledge based radar detection, tracking and classification,” John Wiley and sons, 2008

[9] “Collins paperback English dictionary,” D. Treffy, Ed, Harper-Collins, 2001.

[10] R. L. Gregory “The Osford companion to the mind,” Oxford university press, 1998.

[11] S. Legg and M. Hutter “Universal intelligence: A definition of machine intelligence,” Minds and machines, pp 391-444, Springer, 2007

[12] D. Vernon, G. Metta and G. Sandini “Survey of arteficial cognitive systems: Implications for the autonomous development of mental capabilities in computational elements,’ IEEE Tr on Evolutionary Computation, Vol 11, Issue 2, pp 151-180, April 2007.

1329