Testing of ITU-T G.168 Line-Echo Cancellers using Matlab-GUIDE VLADIMIR MALENOVSKY, ZDENEK SMEKAL Department of Telecommunications Brno University of Technology 118, Purkynova st., 612 00, BRNO CZECH REPUBLIC Abstract: - During the last decade echo cancellers became an integral part of many telecommunication systems commercially available. Testing and standardization of such systems, however, is still under development and there are many aspects which have not been counted in. Above all standardization bodies dealing with line echo and its cancellation stands the ITU-T, which has developed and deployed the G.168 Recommendation. The recommendation itself represents a comprehensive set of tests for echo-canceller evaluation. This article addresses problems which might arise during practical implementation of G.168. It also discusses possibilities how to overcome them using Matlab-GUIDE and other tools. In particular, today’s evaluation systems are real- time hardware systems (embedded systems) which are controlled from a local PC or similarly. They are powerful and efficient but they are also very expensive. If we don’t need a real-time solution and if we accept the fact that the process of echo canceller testing would require much more time, we can do all the work off- line using Matlab. Key-Words: - echo canceller, G.168 recommendation, signal level measurement, testing signals, matlab, implementation 1 Introduction ITU-T G.168 addresses line echoes that arise in circuits called hybrids. This type of echo is known to be relatively short, i.e. 2-4 ms. Nevertheless, it still represents a significant intrusion to speech communication. The majority of systems containing hybrids also employ echo cancellers for efficient elimination of an echo. Such a canceller needs to be inherently adaptive, since a speech signal is known to be non-stationary. The simplest and most popular structure involved in noise cancellation is a transversal filter (FIR) driven by an LMS algorithm. G.168 has been developed to provide a reference standard for echo canceller solutions. It does not dictate the system to be adaptive nor does it require that the algorithm employed must be LMS. It just sets up certain limits and criteria that every echo canceller must meet in order to be compliant with this recommendation. Testing and evaluation of a particular echo canceller according to G.168 involves running a suite of tests with specific testing signals and analysis of measurement results. Every test gives a simple answer of the type PASS/FAIL which answers the question of whether the canceller meets the criteria or not. Development of such testing systems for echo cancellers can be a complicated task. Big companies, such as GL Communications or DSPG Telecom offer complex embedded systems based on single or multiple DSP’s, FPGA for control and glue logic where all testing signals are generated in real- time. They are usually equipped with level measurement device and a graphical user interface. The price of such systems, however, exceeds thousand dollars. Our solution stems from the conception of off- line testing. It means that the testing is done in three consecutive phases (for each test) – generation of testing signals, one-step evaluation and graphical analysis. The first and the third step are accomplished in Matlab. 2 Echo Cancellers The most prevalent cause of echo in PSTN is impedance mismatches within a circuit called hybrid [2]. In this circuit 4-wire phone lines are converted to 2-wire lines and during signal transmission a portion of incoming signal is reflected back to the speaker. This situation is illustrated at Fig. 1. Echo is perceived by the far-end speaker if its level is sufficiently high and if its delay exceeds 10ms.