The role of Signal Processing In Perimeter Intrusion Detection

By Alec Owen.

In their marketing material, perimeter intrusion detection system (PIDS) manufacturers are increasingly using terminology such as ‘advanced signal processing’, ‘intelligent signal processing’, ‘artificial intelligence’ or AI, ‘digital signal processing’ and the like. These terms and the way they are used can be confusing, making it difficult to understand what you are really being offered. The aim of this article is to demystify these term:, what they mean, what they do, and what benefits they bring to the end user.

The Problem

Nuisance alarms are typically generated by a broad range of ‘environmental’ conditions. These can include the wind flapping or rattling the fence fabric; rain or hail falling on the sensors; small wildlife; nearby lightning strikes or thunder; nearby road, rail and air traffic; the list goes on. As sure as death and taxes (and in spite of some manufacturers’ claims!) you will always get some nuisance alarms, but how you deal with them is critical to the overall system performance.

Former Solutions

In the past, PIDS systems had fairly unsophisticated ways of dealing with nuisance alarms. As the background signal level increased – for example as the wind speed increased or the rain got heavier – the sensitivity of the system was reduced to decrease the number of nuisance alarms. The problem is that the ability to detect a real intrusion event is also reduced correspondingly, often to the point of failing to detect a fence climb. So if you were trying to break into a site undetected, the best time to choose would be a wet and windy night.

Setting up and configuring a system was a delicate balancing act, trading off sensitivity against nuisance alarms. How many nuisance alarms per day could the customer tolerate in order to maintain sensitivity? Alternatively, could they eliminate nuisance alarms but live with only detecting
major intrusion events on the perimeter fence?

Some systems needed to be recalibrated each season, some had anemometers connected to them so that as the wind speed picked up the sensitivity of the system was automatically reduced. A range of creative methods have been employed over the years to recognise and eliminate nuisance alarms – some more successful than others. One example was the use of headphones so the alarm operator could ‘listen in’ to the alarm signal to determine if it was a real fence climb or just environmental noise. In this case, the signal discrimination (deciding what is an alarm and what isn’t) is human rather than electronic, and correspondingly (or humanly) highly subjective and inconsistent.

Today

The newer interferometric fibre optic intrusion detection technologies, now widely available on the market, deliver significant improvements in sensitivity when compared to more traditional fibre optic and copper PIDS systems. This improved sensitivity results in a higher probability of detection (POD) of an intruder, especially when they are carefully trying to defeat the system using techniques such as stealthy climbing, carefully propping ladders and even placing ladders with sponge on the back of them against fences in an attempt to climb over undetected. But there is a downside to this – this improved sensitivity can lead to increased nuisance alarms. So the challenge facing manufacturers and developers continues to be how to minimise these nuisance alarms without compromising system sensitivity to real intrusion events.

As these newer systems also cover much longer distances than in the past – typically many kilometers rather than just a few hundred meters per controller – you now have more background noise being detected by a much longer sensor cable in addition to the nuisance alarms generated by the increased sensitivity, so traditional methods of handling environmental noise such as filtering and thresholds will not work. Far more efficient signal processing is required: signal processing that can clearly differentiate between what is a real intrusion and what isn’t.

This is one reason why the newer technologies typically use a centrally housed processor to manage the entire PIDS system. In addition to the substantial installation cost savings and maintenance benefits of not requiring power or communications to the field, one important advantage of this PIDS architecture is that you now have considerable processing ‘horsepower’ readily available. This enables you to implement some very advanced signal processing and identification.

The implementation of these advanced signal processing techniques is transforming the PIDS market. Traditional systems without this level of signal processing will disappear, replaced by these high performance systems that offer greater sensitivity, fewer nuisance alarms, lower overall costs, and much simpler configuration.

The Meaning Behind The Terminology

Notwithstanding the technical definitions, in the PIDS market, the terms ‘advanced signal processing’, ‘intelligent signal processing’, and ‘digital signal processing’ are used loosely and often interchangeably to represent the same thing – using intelligent algorithms to analyse and identify different events within the raw signal. This technology (comprised of both hardware and software) digitally processes the signals received from the fence to remove nuisance events yet retain real intrusion event information.

Artificial intelligence (AI), however, takes this one step further by analysing, classifying, and then comparing the received filtered signal to a known event and actually making the yes or no decision.

Advanced / Intelligent / Digital Signal Processing

These three terms are more generally referred to as Digital Signal Processing or DSP. DSP uses mathematical algorithms rather than traditional analog filtering techniques for processing the raw perimeter sensor signals.

The goal of DSP within a PIDS application is to measure and filter the signals from the sensor and effectively remove those parts of the signal not attributable to a real intrusion event (environmental noise). In most cases this signal processing is a multi-step process. The first step converts the signal from an analog to a digital form as the computational requirements for digital signal processing are far simpler than analog. The signals are converted from time to the frequency domain usually through the Fourier transform, or into the time-frequency domain using wavelet or quadratic time-frequency methods to reveal more information.

The next step is digitally examining the signal to determine which frequencies are present in the signal for a real intrusion event, passing these through, and blocking those frequencies that are known to be caused by environmental and nuisance events.

DSP provides much finer filter control than you could ever achieve with analog components. Any changes required by the filters are done in software rather than in hardware, making them programmable and highly flexible. The downside is the processing overhead required for filtering out bands or frequencies that are not real intrusion events and so are of no interest . For this reason, a growing number of DSP applications are now implemented using powerful head-end PCs with multi-core processors.

Artificial Intelligence

Artificial Intelligence (AI) is the next step in the process after the DSP. AI is so advanced that it builds mathematical models to simulate the human neural decision making processes, replicating in software how your brain makes
a decision.

Neural networks, as used in artificial intelligence, are used to model the complex signals received from the perimeter fence sensor and detect patterns in data. Identifying these patterns in the alarm data is specifically of interest in the quest to eliminate
nuisance alarms.

AI enables the system to recognise and remove background signals such as rain, leaving the intrusion signal untouched without any loss of sensitivity. It then processes this signal further to alarm and locate the intrusion. By employing AI, this nuisance mitigation algorithm dynamically adjusts to varying levels of rain (or other sources of nuisance alarms) but, importantly, never reduces the intrusion event sensitivity.

Only a few years ago, this leading edge technology was confined primarily to the military and aerospace industries. Now it has become an essential part of the technology industry, providing the heavy lifting for many of the
most difficult problems in signal analysis, as
seen in this latest generation of intrusion detection systems.

By definition, AI is an intelligent self-learning process, but in the PIDS industry it is currently implemented at a basic ‘supervised’ level, primarily as a decision making process. There is no doubt that self-learning PIDS systems using ‘unsupervised’ AI methods will appear in the future, but the industry is not there yet.

The most effective PIDS systems currently use a combination of both digital signal processing and artificial intelligence. Digital signal processing does the first pass of the incoming signal to remove those parts of the signal clearly not associated with an intrusion. The remaining signal data is then passed to the artificial intelligence program for further processing that includes features such as signal pattern recognition to provide a more refined level of filtering of nuisance events and the actual decision making. The result is a clear highly accurate and reproducible yes or no intrusion alarm under a range of environmental conditions with very few nuisance alarms.

Alec Owen has almost 20 years experience in advanced sensing technologies and a broad knowledge of the technologies currently employed in outdoor intrusion detection. He is International Client Manager for Future Fibre Technologies, a world-leading designer and manufacturer of fibre optic intrusion detection systems. Alec is the author of The Boundaries of Security – Global Trends in Perimeter Security, a highly respected resource book for the security industry. This book can be downloaded from www.fftsecurity.com

 

%d bloggers like this: