and make capricious course changes. We 

 found it exceedingly difficult to nnaintain con- 

 tact with these schools. Even if contact was 

 not broken the schools had to be kept within 

 100 yards (91.4 m.) of the vessel for Doppler 

 detection, and at these close ranges they 

 usually outmaneuvered us. 



Because of these difficulties we decided to 

 see if the complete target signal, noise Doppler 

 effect and target Doppler effect could be 

 classified by computer. A contract was made 

 with Scope Inc., Falls Church, Va., to explore 

 acoustic methods of classifying characteristics 

 of schooling fish. Specifically, the work was 

 concerned with establishing the feasibility of 

 using adaptive pattern-recognition processes 

 in the automatic classification of Straza CTFM/ 

 CTD sonar returns. 



Data were collected aboard the Bureau of 

 Commercial Fisheries vessel. Miss Behavior , 

 off the coast of southern California, The audio 

 outputs of the CTFM and CTD receivers, as 

 selected in the DELTIC analyzer, were re- 

 corded on one channel of the associated Ampex 

 SP300 magnetic tape recorder; a voice annota- 

 tion was recorded on another channel. When a 

 target searched for in the 200-yard (182.9 m.) 

 and 400-yard (365.8 m.) CTFM mode was 

 located, it was recorded with voice annotation 

 of estimated target size, species, approximate 

 geographical location of the vessel, range-to- 

 target, and sonar mode of operation. Upon 

 target detection, the "Scan Rev" control was 

 manually activated to reduce the scan from 

 ± 45° or 22.5° to about t one-half of the angle 

 subtended by the target. This procedure allowed 

 more nearly constant contact with the target 

 while permitting it to be tracked in azimuth. 



The recordings thus made were subsequently 

 dubbed twice to make them compatible with 

 scope's magnetic-tape reproduction facili- 

 ties. The resulting tape contained six targets 

 and five ambient recordings, totaling about 18 

 minutes of usable data. Table 4 shows a de- 

 tailed breakdown of the data. 



It became apparent after the initialprocess- 

 ing of the magnetic tapes that the limited data 

 base obtained was insufficient for the analysis 

 required to determine the feasibility of using 

 pattern-recognition processes in the automatic 

 classification of sonar returns. We did under- 

 take analysis, however, with the data available 

 to ascertain whether the sonar echo received 

 when a target is present contains anomalies 

 that are not in the echoes received during the 

 recording of ambient conditions. Further, it 

 was intended that, if such anonrialies were 

 found, the analysis would reveal whether they 

 have characteristics that might be used in 

 classifying the target as to size and species. 



The decision to analyze for spectral content 

 the data available in the initial study was 

 dictated by two factors. The first consideration 

 was that the difference frequency in the CTFM 

 mode is the only expression of the information 



available in that mode--target range and rela- 

 tive target motion. The latter produces a 

 Doppler shift of the FM signal, resulting in a 

 modulation of the difference frequency. The 

 second consideration was that (as indicated by 

 an exannination of the physics of the situation) 

 the information present should be in the form 

 of spectral/time distributions. Furthermore, 

 this analysis could fortunately be produced by 

 instrumentation readily available, 



Sonagrams were made of the data on a 

 Sonagraph 606 1 A manufactured by the Kay 

 Electric Company. This spectrunn analyzer 

 produces a frequency versus time versus 

 amplitude (intensity) plot in the 85 Hz to 8 

 kHz range. When a target was known to be 

 present, a signal occasionally appeared at a 

 frequency corresponding to a range approxi- 

 nnating that of the target--when the target 

 range was known. Targets never appeared in 

 more than one -third of the sweeps because 

 of the sector -scanning technique of the sonar. 



In both the CTFM and CTD modes and at 

 all ranges of the CTFM mode a signal appears 

 continuously in the 5 to 5.5 kHz range. Attennpts 

 to correlate the signal with a known target or 

 a natural obstruction such as the ocean floor 

 proved fruitless. 



Further analysis was made on a General 

 Radio Wave Analyzer Model 1900 A and re- 

 corded on a General Radio Graphic Level 

 Recorder Model 1521B. The frequency- versus - 

 amplitude plot produced disclosed a different 

 frequency signal with a value approximating 

 that of the range of the target. Once again, 

 the unidentified signal appeared, this time at 

 5.3 kHz. 



Signals were detected with both the sona- 

 graph and wave analyzer at frequencies ap- 

 proximating those of target range. The signals 

 also appeared at the times when a target was 

 believed to be present, but the detected signal 

 and the target could not be correlated other 

 than in range and time. Classification of the 

 target by species was impossible, and it does 

 not appear practical to look for more informa- 

 tive features until we obtain a larger and 

 more comprehensive data base. 



RECOMMENDATIONS 



To date the sonar experiments have given 

 interesting but inconclusive results. The editor 

 believes that the prospects for commercial- 

 fishing applications of Doppler target classi- 

 fication are at this time financially impractical. 

 The usefulness of the Doppler work in research 

 is less clear. The problem of identifying 

 subsurface sonar contacts for population and 

 behavior studies is yet unsolved. It is too 

 early to say whether acoustic methods --high- 

 resolution, resonant studies or Doppler 

 effects --alone or in combination will prove 

 more satisfactory than other methods such as 



18 



