Analysis of Flight 103 Cockpit Voice Recorder


1. Introduction

This appendix describes and analyses the different types of recorded data which were examined during the investigation of the accident to Boeing 747 registration N739PA at Lockerbie on 21 December 1988. The recorded data consists of that from the Cockpit Voice Recorder (CVR), the Digital Flight Data Recorder (DFDR), Air Traffic Control (ATC) radio telephony (RTF), ATC radar, and British Geological Survey seismic records. The time correlation of the records is also discussed.

2. Digital flight data recorder

The flight data recorder installation conformed to ARINC 573B standard with a Lockheed Model 209 DFDR receiving data from a Teledyne Controls Flight Data Acquisition Unit (FDAU). The system recorded 22 analogue parameters and 27 discrete (event) parameters. The flight recorder control panel was located in the flight deck overhead panel. The FDAU was in the main equipment centre at the front end of the forward hold and the flight recorder was mounted in the aft equipment centre.

2.1 DFDR strip and examination

Internal inspection of the DFDR showed that there was considerable disruption to the control electronics circuits. The crash protection was removed and the plastic recording tape was found detached from its various guide rollers and tangled in the tape spools. There was no tension in the negator springs. This indicated that the tape had probably moved since electrical power was removed from the recorder. The position of the tape in relation to the record/replay heads was marked with a piece of splicing tape in order to quantify the movement. To ensure that no additional damage was caused to the tape it was necessary to cut the negator springs to separate the upper and lower tape reels.

The crinkling and stretching of the tape and the damage to the control electronics meant that the tape had to be replayed outside the recorder. AAIB experience has shown that the most efficient method of replaying stretched Lockheed recorder tapes is to re-spool the tape into a known serviceable recorder, in this case a Plessey 1584G.

2.2 DFDR replay

The 25 hour duration of the DFDR was satisfactorily replayed. Data relating to the accident flight was recorded on track 2. The only significant defect in the recording system was that normal acceleration was inoperative. There was one area on the tape, 2 minutes from the end, where data synchronisation was lost for 1 second.

Decoding and reduction of the data from the accident flight showed that no abnormal behaviour of the data sensors had been recorded. The recorded data simply stopped. Figure C-1 is a graphical representation of the main flight parameters.

2.3 DFDR analysis

In order to ensure that all recorded data from the accident flight had been decoded and to examine the quality of the data at the end of the recording, a section of tape, including both the most recently recorded data and the oldest data (data from 25 hours past), was replayed through an ultra-violet (UV) strip recorder. The data was also digitised and the resulting samples used to reconstruct the tape signal on a VDU.

Both methods of signal representation were used to determine the manner by which the recorder stopped. There was no gap between the most recently recorded data and the 25 hour old data. This showed that the recorder stopped while there was an incoming data stream from the FDAU. The recorder, therefore, stopped because its electrical supply was disconnected. The tape signal was examined for any transients or noise signals that would have indicated the presence of electrical disturbances prior to the recorder stopping. None was found and this indicated that there had been a quick clean break of the electrical supply.

The last seconds of data were decoded independently using both the UV record and the digitised signal. Only 17 bits of data were not recoverable (less that 23 milliseconds) and it was not possible to establish with any certainty if this data was from the accident flight or if it was old data from a previous recording.

A working group of the European Organisation for Civil Aviation Electronics (EUROCAE) was, during the period of the investigation, formulating new standards (Minimum Operational Performance Requirement for Flight Data Recorder Systems, Ref:- ED55) for future generation flight recorders which would have permitted delays between parameter input and recording (buffering) of up to ? second. These standards are intended to form the basis of new CAA specifications for flight recorders and may be adopted worldwide.

The analysis of the final data recorded on the DFDR was possible because the system did not buffer the incoming data. Some existing recorders use a process whereby data is stored temporarily in a memory device (buffer) before recording. The data within this buffer is lost when power is removed from the recorder and in currently designed recorders this may mean that up to 1.2 seconds of final data contained within the buffer is lost. Due to the necessary processing of the signals prior to input to the recorder, additional delays of up to 300 milliseconds may be introduced. If the accident had occurred when tha aircraft was over the sea, it is very probable that the relatively few small items of structure, luggage and clothing showing positive evidence of the detonation of an explosive device would not have been recovered. However, as flight recorders are fitted with underwater location beacons, there is a high probability that they would have been located and recovered. In such an event the final milliseconds of data contained on the DFDR could be vital to the successful determination of the cause of an accident whether due to an explosive device or other catastrophic failure. Whilst it may not be possible to reduce some of the delays external to the recorder, it is possible to reduce any data loss due to buffering of data within the data acquisition unit.

It is, therefore, recommended that manufacturers of existing recorders which use buffering techniques give consideration to making the buffers non-volatile, and hence recoverable after power loss. Although the recommendation on this aspect, made to the EUROCAE working group during the investigation, was incorporated into ED55, it is also recommended that Airworthiness Authorities re-consider the concept of allowing buffered data to be stored in a volatile memory.

3. Cockpit voice recorder (CVR)

The aircraft was equipped with a 30 minute duration 4 track Fairchild Model A100 CVR, and a Fairchild model A152 cockpit area microphone (CAM). The CVR control panel containing the CAM was located in the overhead panel on the flight deck and the recorder itself was mounted in the aft equipment centre.

The channel allocation was as follows:-

Channel 1Flight Engineer’s RTF. Channel 2Co-Pilot’s RTF. Channel 3Pilot’s RTF.Channel 4Cockpit Area Microphone. 3.1 CVR strip and examination

To gain access to the recording tape it was necessary to cut away the the outer case and saw through part of the crash protected enclosure. No damage to the tape transport or the recording tape was found. The endless loop of tape was cut and the tape transferred to the replay equipment. The electronic modules in the CVR were crushed and there was evidence of long term overheating of the dropper resistors on the power supply module. The CAM had been crushed breaking internal wiring and damaging components on the printed circuit board.

3.2 CVR replay

The erase facility within the CVR was not functioning satisfactorily and low level communications from earlier recordings was audible on the RTF channels. The CAM channel was particularly noisy, this was probably due to the combination of the inherently noisy cockpit of the B747-100 in the climb and distortion from the incomplete erasure of the previous recordings. On two occasions the crew had difficulty understanding ATC, possibly indicating high cockpit noise levels. There was a low frequency sound present at irregular intervals on the CAM track but the source of this sound could not be identified as of either acoustic or electrical in origin.

The CVR tape was listened to for its full duration and there was no indication of anything abnormal with the aircraft, or unusual in crew behaviour. The tape record ended with a sudden loud sound on the CAM channel followed almost immediately by the cessation of recording. The sound occurred whilst the crew were copying their transatlantic clearance from Shanwick ATC.

3.3 Analysis of the CVR record

3.3.1 The stopping of the recorder

To determine the mechanism that stopped the recorder a bench test rig was constructed utilizing an A100 CVR and an A152 CAM. Figures C-2 to C-5 show the effect of shorting, earthing or disconnecting the CAM signal wires. Figure C-8 shows the CAM channel signal response to the event which occurred on Flight PA103. From this it can be seen that there are no characteristic transients similar to those caused by shorting or earthing the CAM signal wires. Neither does the signal stop cleanly and quickly as shown in Figure C-5, indicating that the CAM signal wires were not interrupted. The UV trace shows the recorded signal decaying in a manner similar to that shown in Figure C-6, which demonstrates the effect of disconnecting electrical power from the recorder. The tests were repeated on other CVRs with similar results and it is therefore concluded that Flight PA103’s CVR stopped because its electrical power was removed.

Figures C-9A to C-9D show the recorded signals for the Air India B747 (AI 182) accident in the North Atlantic on 23 June 1985. These show that there is a large transient on the CAM track indicating earthing or shorting of the CAM signal wires and that recorder power-down is more prolonged, indicating attempts to restore the electrical power supply either by bus switching or healing of the fault. The Flight PA103 CVR shows no attempts at power restoration with the break being clean and final.

In order to respond to events that result in the almost immediate loss of the aircraft’s electrical power supply it was therefore recommended during the investigation that the regulatory authorities consider requiring CVR systems to contain a short duration (i.e. no greater than 1 minute) back-up power supply.

3.3.2 Information concerning the event

Figure C-8 is an expanded UV trace of the final milliseconds of the CVR record. Three tracks have been used, the flight engineer’s RTF channel which contained similar information to the P2’s channel has been replaced with a timing signal. Individual sections of interest are identified by number. On the bottom trace, the P1 RTF track, section 1 is part of the Shanwick transatlantic clearance. During this section the loud sound on the CAM channel is evident.

Examination of the DFDR event recordings shows that the Shanwick oceanic clearance was being received on VHF2, the aerial for which is on the underside of the fuselage close to the seat of the explosion. Section 2 identifies a transient, on the P1 channel, typical of an end of ATC transmission transient for this CVR. The start and finish of most of the recorded ATC transmissions were analysed and they produce a similar signature to the three shown in Figure C-10. The signature on the P1 channel more closely resembles the end of transmission signature and it is open to conjecture that this transient was caused by the explosion damaging the aerial feeder and/or its supporting structure.

Section 3 shows what is considered to be a high speed power supply transient which is evident on all the RTF channels and is probably on the CAM channel, but cannot be identified because of the automatic gain control (AGC), limiting the audio event. This transient is considered to coincide with the loss of electrical power to the CVR. Section 5 identifies the period to the end of recording and this agrees well with tests carried out by AAIB and independently by Fairchild as part of the AI 182 investigation. The typical time from removal of the electrical supply until end of recording is 110 milliseconds.

During the period identified as section 4 it is considered that the disturbances on the RTF channels are electrical transients probably channelled through the communications equipment. Section 6 identifies the 170 millisecond period from the point when the sound was first heard on the CAM until the recording stopped.

The CAM unit is of the old type which has a frequency response of 350 to 3500 Hz. The useable duration of the signal is probably confined to the first 60 milliseconds of the final 170 milliseconds and even during this period the AGC is limiting the signal. In the remaining time the sound is being distorted because power to the recorder has been disconnected. The ambient cockpit noise may have been high enough to have caused the AGC to have been active prior to the event and in this event the full volume of the sound would not be audible. Distortion from the incomplete erasure of the last recording may form part of the recorded signal.

It is not clear if the recorded sound is the result of the explosion or is from the break-up of the aircraft structure. The short period between the beginning of the event and the loss of electrical power suggests that the latter is more likely to be the case.

Additionally some of the frequencies present on the recording were not present in the original sound, but are the result of the rise in total harmonic distortion caused by the increased amplitude of the incoming signal. Outputs from a frequency analysis of the recorded signal for the same frequency of input to the CVR, but at two input amplitudes, are shown in Figures C-11 and C-12. These illustrate the effects on harmonic distortion as the signal level is increased. Finally the recorded signal does not lend itself to analysis by a digital spectrum analyser as it is, in a large measure, aperiodic and most digital signal analysis algorithms are unable to deal with a short duration signal of this type, however, it is hoped that techniques being developed in Canada will enable more information to be deduced from the end of the recording.

In the aftermath of the Air India Boeing 747 accident (AI 182) in the North Atlantic on 23 June 1985 the Royal Armaments Research and Development Establishment (RARDE) were asked informally by AAIB to examine means of differentiating, by recording violent cabin pressure pulses, between the detonation of an explosive device within the cabin (positive pulse) and a catastrophic structural failure (negative pulse). Following the Lockerbie disaster it was considered that this work should be raised to a formal research project. Therefore, in February 1989, it was recommended that the Department of Transport fund a study to devise methods of recording violent positive and negative pressure pulses, preferably utilising the aircraft’s flight recorder systems.

Preliminary results from these trials indicates that if a suitable sensor can be developed its output will need to be recorded in real time and therefore it may require wiring into the CVR installation. This will further strengthen the requirement for battery back up of the CVR electrical power supply.

4. Flight recorder electrical system

4.1 CVR/DFDR electrical wiring.

The flight recorders were located in the left rear fuselage just forward of the rear pressure bulkhead. Audio information to the CVR ran along the left hand side of the aircraft, at stringer 11. Electrical power to the CVR followed a similar route on the right hand side of the aircraft crossing to the left side above the rear passenger toilets. DFDR electrical power and signal information followed the same route as the CVR audio information.

4.2 Flight recorder power supply

The DFDR, CVR and the transponders were all powered from the essential alternating current (AC) bus. This bus was capable of being powered by any generator, however, in normal operation the selector switch on the flight engineers panel is selected to "normal" connecting the essential bus to number 4 generator. When the cockpit of Flight PA103 was examined the selector switch was found in the normal position.

4.3 Aircraft alternating current power supplies

AC electrical power to the aircraft was provided by 4 engine driven generators, see Figure C-13. Each generator was driven at constant speed through a constant speed drive (CSD) and connected to a separate bus-bar through a generator control breaker (GCB). The 4 generators were connected to a parallel bus-bar (sync bus) by individual bus tie breakers (BTBs). Control and monitoring of the AC electrical system was achieved through the flight engineer’s instrument panel. In normal operation the generators operated in parallel, i.e with the BTBs closed.

4.4 Fault conditions

Analysis of the CVR CAM channel signal indicated that approximately 60 milliseconds after the sound on the CAM channel an electrical transient was recorded on all 4 channels and that approximately 110 milliseconds later the CVR had ceased recording. Within the accuracy of the available timing information it is believed that the incoming VHF was lost at the same time, indicating an AC power supply fault.

The AC electrical system was protected from faults in individual systems or equipment by fuses or circuit breakers. Faults in the generators or in the distribution bus-bars and feeders were dealt with automatically by opening of the GCBs and opening or closing of the BTBs. In the event of fault conditions causing the disconnection of all 4 generators electrical power for essential services, including VHF radio, was provided by a battery located in the cockpit.

The short time interval of 55 milliseconds after which the AC supply to the flight recorders was lost limits the basis on which a fault path analysis of the AC electrical system can be undertaken. On the available information only a differential (feeder) fault could have isolated the bus-bar this quickly, with the generator field control relay taking 20 milliseconds to trip. However, in normal operation, the generators would have been operating in parallel and the essential AC bus-bar would have been supplied via the number 4 BTB from the sync bus. If the fault conditions had continued, a further 40 to 100 milliseconds would have elapsed before the BTB opened. If the BTB was open prior to the fault it would have attempted to close and restore the supply to the essential bus. Any automatic switching causes electrical transients to appear on the CVR and data losses on the FDR. Both the CVR and the FDR indicate that a clean break of the AC supply occurred with no electrical transients associated with BTBs open or closing in an attempt to restore power. In the absence of any additional information only two possibilities are apparent:

i) That all 4 generators were simultaneously affected causing a total loss of AC electrical power. The feeders for the left and right side generators run on opposite sides of the aircraft under the passenger cabin floor. The only situation envisaged that could cause simultaneous loss of all 4 generators is the disruption of the passenger cabin floor across its entire width.

ii) That disruption of the main equipment centre, housing the control units for the AC electrical system, caused the loss of all AC power. However, again it would have to affect both the left and right sides of the aircraft as the control equipment is located at left and right extremes of the main equipment centre.

The nature of the event may also produce effects that are not understood. It is also to be noted that a sudden loss of electrical power to the flight recorders has been reported in other B747 accidents, e.g. Air India, AI 182.

5. Seismic data

The British Geological Survey has a number of seismic monitoring stations in Southern Scotland. Stations close to Lockerbie recorded a seismic event caused by the wing section crashing on Lockerbie. The seismic monitors are time correlated with the British Telecom Rugby standard. Using this and calculating the time for the various waves to reach the recording stations it was possible for the British Geological Survey to conclude that the event occurred at 19.03:36.5 hrs ? 1 second.

Attempts were made to correlate various smaller seismic events with other wreckage impacts. However, this was not conclusive because the nearest recording station was above ground and due to the high winds at the time of the accident had considerable noise on the trace. In addition, little of the other wreckage had the mass or impact velocity to stimulate the sensors.

6. Time correlation

6.1 Introduction

The sources of each time encoded recording were asked to provide details of their time standard and any known errors in the timings on their recordings. Although the resolution of the recorded time sources is high it was not possible to attach an accuracy of better than ?1 second due to possible errors in synchronising the recorded time with the associated standard. The following time sources were available and used in determining the significant events in the investigation:-

i) ATC

ATC communications were recorded along with a time signal. The time source for the ATC tape was the British Telecom "Tim" signal. Any error in setting the time when individual tapes are mounted was logged.

ii) Recorded rada data

A time signal derived from the British Telecom "Rugby" standard was included on radar recordings. The Rugby and Tim times were assumed to be of equal accuracy for timing purposes.

iii) The DFDR had UTC recorded.

The source of this time was the flight engineer’s clock. This clock was set manually and therefore this time was subject to a significant fixed error as well any inaccuracy in the clock.

iv) The CVR had no time signal.

However, the CVR was correlated with the ATC time through the RTF and with the DFDR, by correlating the press to talk events on the FDR with the press to talk signature on the CVR.

v) Seismic recordings

Seismic recordings included a timing signal derived from the British Telecom Rugby standard.

6.2 Analysis and correlation of times

The Scottish and Shanwick ATC tapes were matched with each other and with the CVR tape. The CVR recording speed was adjusted by peaking its recorded 400 Hz AC power source frequency. This correlation served as a double check on any fixed errors on the ATC recordings and to fix events on the CVR to UTC. The timing of the sound on the CAM channel of the CVR was made simpler because Shanwick was transmitting when it occurred. From this it was possible to determine that the sound on the CVR occurred at 19.02:50 hrs ?1 second.

With the CVR now tied to the Tim standard it was possible to match the RTF keying on the CVR with the RTF keying events on the FDR. These events on the FDR were sampled and recorded once per second, it was therefore possible for a 1 second delay to be present on the FDR. This potential error was reduced by obtaining the best fit between a number of RTF keyings and a time correlation between the FDR and CVR of ?? second was achieved. From this it was determined, within this accuracy, that electrical power was removed from the CVR and FDR at the same time.

From the recorded radar data it was possible to determine that the last recorded SSR return was at 19.02:46.9 hrs and that by the next rotation of the radar head a number of primary returns, some left and right of track, were evident. Time intervals between successive rotations of the radar head became more difficult to use as the head painted more primary returns.

The point at which aircraft wreckage impacted Lockerbie was determined using the time recorded by seismic activity detectors. A seismic event measuring 1.6 on the Richter scale was detected and, with appropriate time corrections for times of the waves to reach the sensors, it was established that this occurred at 19.03:36.5 hrs ?1 second. A further check was made by triangulation techniques from the information recorded by the various sensors.

7. Recorded radar information

7.1 Introduction

Recorded radar information on the aircraft was available from from 4 radar sites. Initial analysis consisted of viewing the recorded information as it was shown to the controller on the radar screen, from this it was clear that the flight had progressed in a normal manner until Secondary Surveillance Radar (SSR) was lost. There was a single primary return received by both Great Dun Fell and Claxby radars approximately 16 seconds before SSR returns were lost. The Lowther Hill and St. Annes radars did not see this return. The Great Dun Fell radar recording was watched for 1 hour both before and after this single return for any signs of other spurious returns, but none was seen. The return was only present for one paint and no explanation can be offered for its presence.

7.2 Limitations of recorded radar data

Before evaluating the recorded radar data it is important to highlight limitations in radar performance that must be taken into account when interpreting primary radar data. The radar system used for both primary and secondary radar utilised a rotating radar transmitter/receiver (Head). This means that a return was only visible whilst the radar head was pointing at the target, commonly called painting or illuminating the target. In the case of this accident the rotational speeds of the radar heads varied from approximately 10 seconds for the Lowther Hill Radar to 8 Seconds for the Great Dun Fell Radar.

Whilst it was possible to obtain accurate positional information within a resolution of 0.09? of bearing and ? 1/16 nautical mile range for an aircraft from SSR, incorporating mode C height encoding, primary radar provided only slant range and bearing and therefore positional information with respect to the ground was not accurate.

The structural break-up of an aircraft releases many items which were excellent radar reflectors eg. aluminium cladding, luggage containers, sections of skin and aircraft structure. These and other debris with reflective properties produce "clutter" on the radar by confusing the radar electronics in a manner similar to chaff ejected by military aircraft to avoid radar detection.

Even when the target is not masked by clutter repetitive detection of individual targets may not be possible because detection is a function of the target effective area which, for wreckage with its irregular shape, is not constant but fluctuates wildly. These factors make it impossible to follow individual returns through successive sweeps of the radar head.

7.3 Analysis of the radar data

The detailed analysis of the radar information concentrated on the break-up of the aircraft. The Royal Signals and Radar Establishment (RSRE) corrected the radar returns for fixed errors and converted the SSR returns to latitude and longitude so that an accurate time and position for the aircraft could be determined. This information was correlated with the CVR and ATC times to establish a time and position for the aircraft at the initial disintegration.

For the purposes of this analysis the data from Great Dun Fell Radar has been presented. Figures C-14 to C-23 show a mosaic picture of the radar data i.e. each figure contains the information on the preceding figure together with more recently recorded information. Figure C-14 shows the radar returns from an aircraft tracking 321?(Grid) with a calculated ground speed of 434 kts. Reading along track (towards the top left of Figure C-14) there are 6 SSR returns with the sixth and final SSR return shown decoded: squawk code 0357 (identifying the aircraft as N739PA); mode C indicating FL310; and the time in seconds (68566.9 seconds from 00:00, i.e. 19.02:46.9 hrs).

At the next radar return there is no SSR data, only 4 primary returns. One return is along track close to the expected position of the aircraft if it had continued at its previous speed and heading. There are 2 returns to the left of track and 1 to the right of track. Remembering the point made earlier about clutter, it is unlikely that each of these returns are real targets. It can, however, be concluded that the aircraft is no longer a single return and, considering the approximately 1 nautical mile spread of returns across track, that items have been ejected at high speed probably to both right and left of the aircraft. Figure C-15 shows the situation after the next head rotation. There is still a return along track but it has either slowed down or the slant range has decreased due to a loss of altitude.

Each rotation of the radar head thereafter shows the number of returns increasing with those first identified across track in Figure C-14 having slowed down very quickly and followed a track along the prevailing wind line. Figure C-20 shows clearly that there has been a further break-up of the aircraft and subsequent plots show a rapidly increasing number of returns, some following the wind direction and forming a wreckage trail parallel to and north of the original break-up debris. Additionally it is possible that there was some break-up between these points with a short trail being formed between the north and south trails. From the absence of any returns travelling along track it can be concluded that the main wreckage was travelling almost vertically downwards for much of the time.

The geographical position of the final secondary return at 19.02:46.9 hrs was calculated by RSRE to be OS Grid Reference 15257772, annotated Point A in Appendix B, Figure B-4, with an accuracy considered to be better than ?300 metres This return was received 3.1?1 seconds before the loud sound was recorded on the CVR at 19.02:50 hrs. By projecting from this position along the track of 321?(Grid) for 3.1?1 seconds at the groundspeed of 434 kts, the position of the aircraft was calculated to be OS Grid Reference 14827826, annotated Point B in Appendix B, Figure B-4, within an accuracy of ?525 metres. Based on the evidence of recorded data only, Point B therefore represents the geographical position of the aircraft at the moment the loud sound was recorded on the CVR.

8. Conclusions

The almost instant destruction of Flight PA103 resulted in no direct evidence on the cause of the accident being preserved on the DFDR. The CVR CAM track contained a loud sound 170 milliseconds before recording ceased. Sixty milliseconds of this sound were while power was applied to the recorder; after this period the amplitude decreased. It cannot be determine whether the decrease was because of reducing recorder drive or if the sound itself decreased in amplitude. Analysis of both flight recorders shows that they stopped because the electrical supply was removed and that there were valid signals available to both recorders at that time.

The most important contribution to the investigation that the flight recorders could make was to pinpoint the time and position of the event. As the timescale involved was so small in relation to the resolution and accuracy of many of the recorded time sources it was necessary to analyse collectively all the available recordings. From the analysis of the CVR, DFDR, ATC tapes, radar data and the seismic records it was concluded that the loud sound on the CVR occurred at 19.02:50 hrs ?1 second and wreckage from the aircraft crashed on Lockerbie at 19.03:36.5 hrs ?1 second, giving a time interval of 46.5 ?2 seconds between these two events. When the loud sound was recorded on the CVR, the geographical position of the aircraft, based on the evidence of recorded data, was calculated to be within 525 metres of OS Grid Reference 14827826.

Eight seconds after the sound on the CVR the Great Dun Fell radar showed 4 primary radar returns. The returns indicated a spread of wreckage in the order of 1 nautical mile across track. On successive returns of the radar, two parallel wreckage trails are seen to develop with the second trail, to the north, becoming evident 30 to 40 seconds after the first.