Spectral Autonomy

Electronic attack, protection, and cognitive radar

Part of the Field Guide to Cognitive Electronic Warfare.

Electronic attack and electronic protection concern opposing uses of the electromagnetic environment. Cognitive methods introduce feedback and learning into those functions, making their behavior depend on both the received evidence and the responses of other systems.

In this guide

When the system being observed also adapts

A receiver can learn from a sequence of observations while the source of those observations is changing its own behavior. This makes the task different from recognizing recordings in a fixed dataset. A decision by one system changes the conditions encountered by another, whose response may then change the first system's next observation.

Electronic attack and electronic protection bring this interaction into view. Electronic attack seeks to impair an opposing use of the electromagnetic spectrum. Electronic protection preserves the operation of friendly systems despite interference and other adverse conditions. Both can use feedback, but they act toward different objectives and observe different parts of the interaction.

To understand the engineering problem, consider two systems sharing a changing signal environment. One adjusts how it observes or communicates in response to interference. The other then receives a different pattern of signals and revises its own estimate. A model learned before the first adjustment may no longer describe the behavior being observed afterward.

This is coupled adaptation: each system's changes can affect the evidence available to the other. The figure shows those dependencies without prescribing a particular method or operating sequence.

Two systems in a shared electromagnetic environmentEach system receives observations from and contributes behavior to a shared environment. Each has its own objective and model.System AObjective + modelSystem BObjective + modelShared electromagnetic environment

Scroll horizontally to inspect the diagram.

Figure 1. Both systems act on a shared electromagnetic environment. A change in one system's behavior can alter the observations on which the other bases its next decision.

Radar provides a sensing example

Radar transmits electromagnetic energy and uses the returned signal to estimate properties of its surroundings. A radar tracking a moving object, for example, updates its estimate as new returns arrive. Its next sensing decision can be informed by what is already known and by where additional information would be useful.

In cognitive radar, this connection is described as a perception–action cycle. The system uses received evidence to form an estimate, selects a subsequent sensing action, and incorporates the resulting observation. Haykin's account connects this cycle with memory and learning, so that accumulated experience can influence future sensing decisions.

The example also explains why an external observer faces an inference problem. A change in the radar's observable behavior may follow a change in the scene, a revised estimate, or a response to interference. The observer sees the resulting signal but does not necessarily have the internal information that caused the radar's decision. Understanding the behavior requires a model of those possible causes.

What changes when a system adapts

Adaptation can occur at several levels. A sensing system may select one of its existing observation modes. It may keep the same mode but adjust a setting within it. A learning method may instead change the rule used to choose between future observations. These changes affect different parts of the system and should be described separately.

The distinction matters when assessing an improvement. A method that selects among existing modes can be compared with another selector given the same choices. A method that changes the decision rule needs an evaluation showing how the rule changes with experience and what difference that makes to later behavior. The test follows the proposed mechanism.

Inferring another system's purpose

An observer may also try to infer what another system is attempting to accomplish. A sequence of sensing decisions can contain evidence about its priorities, but an action has meaning only in relation to the choices and constraints available to that system.

For example, repeated observation of one region could reflect interest in that region, uncertainty about its contents, or a limitation that prevents observations elsewhere. The same visible choice can therefore arise from different internal circumstances. A model that estimates objectives must account for those alternatives to the extent that the available evidence permits.

Such an estimate can still be useful. It organizes possible explanations and predicts which observations would be consistent with each one. As later evidence arrives, the observer can revise the estimate. The result remains an inference about purpose, connected to a stated model and observation history.

Testing an interaction over time

A test with a fixed signal source can establish how a component responds to that source. Studying coupled adaptation requires the test to allow the observed behavior to change as well. Otherwise, the experiment omits the feedback that defines the interaction.

The evaluation record should connect each system's observations, decisions, and changes over time. That record helps explain whether an apparent improvement came from learning, a more favorable input, or a change elsewhere in the experiment. When learned components are involved, their model states and update data are part of the record. NIST's risk-management framework provides broader guidance on documenting, measuring, and monitoring these dependencies.

The final chapter follows this evidence into an integrated capability, where processing delays, interfaces, and operator interaction also affect the result.