Introduction

The idea of machine recognition of Morse code is not a new one. Much work has been done to develop equipment for the automatic reception of Morse code signals in the last two decades. The first notable effort was undertaken by Lincoln Laboratories at the Massachusetts Institute of Technology in the late 1950s (Ref 4). The machine which resulted from this project was actually a special purpose digital computer, called MAUDE (Morse Automatic DEcoder). The recognition algorithm used in MAUDE was based on its “knowledge” of Morse code linguistic properties and of the relative time durations of marks (referred to as pulses throughout this report) and spaces. MAUDE demonstrated a 90 to 95 per cent correct decoding rate, although this rate was later improved by the addition of an output error detection and correction scheme. The main disadvantage of MAUDE» however, was its physical size and complexity.

The development of small, low-cost integrated circuits led to the design and faurication of several Morse-to-teletype converters. Generally speaking, much simpler code recognition algorithmsr as compared to those used in MAUDE, were used in these converters to identify pulses and spaces. One such converter, designed at the Naval Postgraduate School in 1968 (Ref 6), uses the time duration of the most recently received short pulse (DOT) as a reference unit for pulse and space identification. Multiples of this reference unit are used as decision thresholds to identify succeeding pulses and spaces, and to eliminate signal noise.

Another example of this type of converter, called the Morse-A-Verter, uses a variation of the previous recognition algorithm to make pulse-space decisions (Ref 5). The time duration of the most recently processed pulse, short or long, is used to determine the classification of the succeeding pulse. Spaces are identified by comparison with the time duration of tne most recent long pulse (DASH).

Morse code recognition machines perform five basic functions. First, the pulse-modulated audio Morse signal is converted into a form usable by the machine, usually dc pulses, while discriminating against noise. Second, the time duration and identification of each pulse and space is determined. Third, pulses and spaces are classified into one of two pulse or three space categories according to their relative time duration. Fourth, the categorized pulses and spaces are combined to form Morse code characters. Finally, a signal representing the identified Morse code character is transmitted to an output device.

The third function, that of categorizing pulses and spaces, is the most difficult to perform. This difficulty is mainly due to the inherently non-uniform pulse and space time durations of hand-sent code. Since these time durations form the basis for the recognition process, a rigid set of decision algorithms, such as those used in commercial Morse telegraphy equipment, cannot be used. Instead, algorithms based on traits common to all variations of hand-sent Morse code must be used to achieve the highest possible degree of machine recognition accuracy.

The objective of this study is two fold; first, to conduct a thorough examination of hand-sent Morse code data and identify common traits which may be used in a machine recognition process, and second, to develop a Morse code recognition program, for use on the PDP-12 digital computer» to test and refine decision algorithms based on these common traits. Thus» the main concern here is optimization of the third function performed by Morse code recognition machines» as previously defined. Of course» all five functions must be considered in the development of the PDP-12 computer program. The remaining four functions» however» are of secondary concern to this project.

A discussion on the properties of hand-sent Morse code, and on the primary factors responsible for the variations present in hand-sent Morse code» is presented in Chapter II. Chapter III describes the procedure used to obtain and analyze Morse code data» and the common traits discovered during and the decision algorithms derived from this analysis procedure. A complete description of the resulting computer recognition program is presented in Chapter IV» as well as a brief description of the PDP-12 computer and peripheral devices used in this project. The operational procedure used with the recognition program is described in Chapter V. Chapter VI presents an analysis of the results obtained during the testing procedure. Conclusions and recommendations are contained in Chapter VII.