This paper was originally presented at EMC Zurich `95, 11th International Zurich Symposium on EMC, 7-9 March 1995.
In mid-1994 a case was heard at the High Court in London in which the central issue before the court was the electromagnetic susceptibility of an electronic system. The paper discusses issues raised in argument during the trial and in the judgement, as they relate to developments in the wider EMC field. Particular attention is paid to
In the early 80s a manufacturer of fire alarm systems designed an innovative new system that was a first for the fire alarm industry in its communication and fault-tolerance capabilities. Over the next few years the system was installed in several large public venues as well as many smaller buildings. In 1988 the company was bought from its parent by another company looking to break into the fire alarm market. The sale was accompanied by a warranty which stated among other things that the company's designs were free from defect.
Some time after the takeover, the company's new owners noticed that systems in some locations seemed to require an unusually high number of maintenance call-outs, although other locations were trouble-free. After spending a fair amount of development effort in attempting to trace the problem, the design engineers of the second company - who of course were not the system's original designers - concluded that the technique used for communicating between the main control panel and the outlying sensors was inherently susceptible to noise. Several modifications were made to the circuits in an attempt to improve the situation, although these did not include the obvious and straightforward step of filtering the interfaces. Eventually, the purchaser decided that the system's design was fundamentally flawed by virtue of this susceptibility, and that the warranty of freedom from defect was therefore false, and so began proceedings against the vendor for breach of warranty.
During the course of preparation for the trial, the plaintiff submitted a sample of the system to an independent accredited EMC test house and asked them to prepare a report of their findings. The test house applied the tests of IEC801 parts 3 and 4 (RF and fast transient burst immunity) as well as inspecting the design and construction of the system's main control panel. Their report criticized the design for its lack of EMC protective measures as well as claiming that the unit failed several of the tests.
During the trial, the most important points raised by the plaintiff were:
The original designer took several specific precautions in the system, particularly with regard to the operating software, to ensure that it was suitably robust to the anticipated interference. During cross-examination though, he was unable to show how he had quantified either the environment or the expected coupling mechanisms. The argument ran as follows: if he did not know, quantitatively, how interference would be coupled into the system, then it was impossible for him to have properly designed the system to have avoided susceptibilities. At the time of the design - around 1983 - little information was available regarding expected amplitudes, waveshapes and coupling of transients, or indeed regarding the electromagnetic environment in general. Reference was made by the plaintiff to General Electric's Transient Voltage Suppression Manual [1] published in 1978. It is this author's opinion that although that handbook was available around that time, it would have been unusual to find designers of fire alarm systems having it, understanding its contents and using it regularly as an integral part of the design process.
Nevertheless, this view did not find favour with the court. The judgement made clear that the designer was at fault, not only in not designing adequately to cope with the expected environment, but also in not re-designing and upgrading the system when it appeared to be suffering susceptibility problems.
The judge said that "steps could and should have been taken in the design of the printed circuit board to reduce its vulnerability to noise". The PCB layout was typical of early 1980's microprocessor designs - double sided board with no ground plane, and the ground routing included some longer-than-desirable loops. Decoupling was adequate but not in line with guidance that began to appear in manufacturers' data in the mid-80s. According to the judgement, not redesigning the product to take these guidelines into account was a failing for which the original company could be held responsible.
Clearly, this has serious implications for other companies who have fielded designs which do not implement current best practice EMC principles. Apparently it will not be an adequate defence, in the event of a court challenge to the electromagnetic immunity and hence fitness for purpose of such a design, to say that such principles were less well understood when the product was designed. The judgement quoted a passage from a book by the author of this paper [2] which ran as follows:
Having advised segregation of different cable classes, it is still true that the best equipment design will be one which puts no restrictions on cable routing and mixing - i.e. one where the major EMC design measures are taken internally. There are many application circumstances when the installation is carried out by unskilled and untrained technicians who ignore your carefully specified guidelines, and the best product is one which works even under these adverse circumstances.
The fact that this passage advises on best practice, not on mandatory minima, was not given great weight. In the context of a fire alarm system, whose purpose is safety-of-life related, best practice is evidently regarded as mandatory.
The judge had further criticisms of the EMC testing that was carried out before the product was launched. No standard tests were available at that time, and the designers used an ad hoc method, which to them appeared to be relevant to the expected environment of a fire alarm system. They applied radiated noise to the communications wiring from a nearby unsuppressed fire alarm bell. This approach was said by the judge to be not "sufficiently scientific", since no measurements of the actual level of applied interference were made.
Of course, since that time the fast transient burst test of IEC801-4 [3] was established. Since this test was subsequently invoked in a revision of the British Standard for fire alarm control equipment (BS5839 part 4 : 1988) [4] the plaintiff argued that the designers should have applied this test as soon as it became publicly known, that is when the draft of the new BS5839 was circulated. The defendant company had a representative on the appropriate committee and would therefore have known of the impending requirement. In the event, the judgement made no reference to this aspect of the case. Again, the implications for retrospective testing of products when standards are revised are severe: as soon as a new version is published which calls for extra testing, manufacturers would have to rush to apply the tests to all of their products in order to remain within the terms of the standard and hence free from liability. Needless to say, this is not common practice at present.
The purpose of testing in the context of a court case is to lend weight to your own arguments or to rebut the other side's. Litigation "experiments-in-chief" are carried out to establish scientific "facts" which may be relied upon at trial. The classical format for "experiments-in-reply" is to effectively repeat the opponent's experiment using his own techniques but with some apparently minor variant which yields a quite different result [5]. Litigation experiments of this sort are well understood in many fields of industry and especially in patent actions, but because of the lack of cases they are not familiar to the EMC community. One of the effects of the EMC Directive, as well as cases such as that under discussion, will undoubtedly be to increase the number of such "experiments" that test houses are called upon to perform.
Are test houses well prepared for this task? The first and most important requirement is that the tests must be impeccably documented. Every detail of the test which could affect the result - and of course, in EMC testing there are many such details - must be recorded in such a way that the evidence can be produced and defended in court. Test report formats and quality vary enormously between test houses, and considerable discipline will be needed to draft reports which can withstand the intense scrutiny to which they and their authors will be submitted during the trial. In essence, these reports are presented as evidence of fact, and as such are subject to cross-examination by the opposing side's counsel. Any discrepancies will be ruthlessly exposed.
Unfortunately the well-known sensitivity of EMC tests to apparently minor variations - that is, minor in the context of typical installation practice, such as layout, length and type of cables - means that it is likely that both experiments in chief and experiments in reply will be equally valid yet produce markedly different results. Also, it will be possible for litigants to argue that tests done in different test houses (for opposing parties) are not comparable, because of such detailed differences. Undoubtedly the work underway in IEC, CISPR and CENELEC to improve the repeatability of EMC tests will help the situation in the longer term.
At the trial, there was considerable argument over whether the results presented by the test house on behalf of the plaintiff actually showed the equipment to have failed the tests. The standards are open to interpretation both over the application of the tests - i.e. what product configuration should be tested - and over the choice of criterion for failure. Normally, these factors are agreed between the manufacturer and the test house as part of the test plan. In this case, of course, the defendant was presented with a fait accompli and was able to argue, with some degree of success, that the parameters had been chosen unfairly. For instance, the test configuration chosen by the plaintiff was based on a worst case cable type which, though not strictly forbidden by the installation instructions, was not generally used in actual installations. Therefore, it was argued, test results on this cable type were not directly applicable to the question of the system's susceptibility as installed.
... No degradation of performance or loss of function is allowed below a performance level specified by the manufacturer, when the apparatus is used as intended. In some cases the performance level may be replaced by a permissible loss of performance. If the minimum performance level or the permissible performance loss is not specified by the manufacturer then either of these may be derived from the product description and documentation and what the user may reasonably expect from the apparatus if used as intended.
By contrast, clause 15.3.1 of the fire alarm standard (BS5839 part 4) [4] is a model of brevity:
The specimen shall remain in the normal condition during the conditioning period, apart from alarm or fault signals of a purely transitory nature, except when being subjected to a functional test.
When under test, the system exhibited several indications that came and went while the test bursts were applied. Considerable court time was spent arguing whether the observed effects under test fell within or outside the definition of "signals of a purely transitory nature". Had the looser definition of the generic standard been invoked, even more arguments would undoubtedly have been deployed. For instance, what may the user reasonably expect? What performance loss is permissible? In what cases may this be accepted? How is it to be derived from the documentation? These arguments, it seems, are going to be inevitable in any case where there is dispute over the outcome of EMC tests, simply because the standards are necessarily drafted in very vague terms. In the end, it is the judge who must decide on the interpretation of the standards. The need for clear and detailed criteria to be included in the evolving product-specific standards is evident.
With respect to the reliability of methods of test, there is a heavy responsibility on those drafting international EMC standards. Once a test method has been written into an IEC or CISPR standard, that test is then regarded as impeccable by those non-technical people, i.e. the judiciary, who are called upon to judge its results. The EMC community knows that the test methods published in IEC or CISPR standards are necessarily compromises, and are necessarily evolving as knowledge of their weaknesses develops. The judiciary do not know this. Although accomplished at adversarial argument, they are (by their own admission) technically naive, and must rely on experts employed by both sides to illuminate the technical issues. If the experts do not agree - and of course there is much room for disagreement in considering EMC test methods - then the court must fall back on the provenance of the cited authorities. As a result, it is virtually impossible to argue in court that shortcomings in a test method mean that the results of the test should be treated with caution and should not necessarily be seen as a final arbiter. If the tests were performed in compliance with international standards, then their results are regarded as unassailable.
An example of this dilemma is the question of how representative of the real environment is the test. In the case discussed above, the court had to decide whether the system in question actually was affected by environmental electromagnetic interference, and the test results to IEC801 part 4 were cited by the plaintiff as evidence that it was. The transient bursts applied during this test are specified as a frequency of 5kHz, applied with a repetition period of 300ms. These characteristics of the applied interference defeated the software filtering of the system's communication mechanism, which was designed in anticipation of random bursts of interference and which used a signalling frequency very close to 5kHz. Appendix A of IEC801 part 4 [3] in fact states
The actual phenomenon of a burst occurs with repetition rates of the individual pulses from 10kHz to 1MHz. However, extensive investigations have indicated that this relatively high repetition rate is difficult to duplicate with a generator using a fixed adjusted spark gap. Therefore lower repetition rates (but of representative individual pulses) have been specified.
The defendant therefore argued strenuously that, because of the limitations of the available test method, the test of IEC801 part 4 did not properly represent the real environment; indeed, para 7.4.4. of the American standard IEEE C62.41 [7] explicitly states of the fast transient burst test that
it is not an attempt to reproduce the surges as they appear on the mains interface, as other surge waveforms do; it is a practical compromise for evaluating equipment immunity to fast transients.
It was also argued that the parameters chosen for the repetition rate and frequency in IEC801 part 4, because they coincided with system design parameters, caused the test to be unnecessarily stressful on the system. None of this was acceptable to the judge, who took the view that the avowed purpose of the standard was to test against transient interference, the standard was the correct one to apply to this type of equipment, and therefore that it was entirely appropriate to use results obtained from tests done under this standard.
All this may be right and proper, and it is undoubtedly the case that standard test methods are needed and that the IEC series is the correct "home" for such standards. Nevertheless, a clearer exposition within the body of, or as an annex to, the standard, both of the rationale for deciding on a test method and on its limitations, would be highly desirable. This would allow a more informed judgement of the applicability of standard tests to any given case, when the standard is invoked. Goedbloed has argued forcefully [8] that basic EMC standards should be provided with a bibliography referring to technical information supporting the various decisions made and making them traceable. He defines "traceability" as
the property of an EMC test, according to which the rationale of each test item can be related to open references, usually international reports and publications.
An example of such a practice is found in C62.41 [7], quoted above, which contains both an extensive bibliography and an extensive discussion of the rationales for various tests. Such material is generally absent from IEC standards. Although this does not affect the use of the test in straightforward cases of declaring compliance to standard requirements, it severely restricts the utility of the document - as well as the rigour of judgements that are made on its basis - when questions are raised as to its validity and applicability. These questions will undoubtedly be raised in the context of court cases such as that described above, for as long as the tests are subject to a hard-to-quantify uncertainty and for as long as the test methods are less than fully matched to the environmental conditions they seek to represent. These considerations will probably apply to EMC test standards for the foreseeable future. Such questions of applicability and validity should be capable of being resolved fully from within the document, so that there is no need to bring into the argument opinions from external authorities, whose standing the court may not be best placed to judge.
Industry should be aware that its duty to provide products that are fit for their intended purpose extends to ensuring their immunity from electromagnetic interference in whatever environment they may expect to be installed. Any restrictions that this may require in installation practice should be made clear to the installer and purchaser. Constant vigilance is needed to ensure that products keep pace with changes in standard requirements.
Test houses must be prepared to expend considerable effort in documenting tests that they carry out which will subsequently be presented in court. Much greater care and detail is needed than is usual for presenting a customer with a simple statement of pass or failure.
Greater circumspection is needed in drafting EMC testing standards that will be used in court, especially with the promise of actions under the EMC Directive. At the very least, some warning should be embodied in the standard regarding the uncertainties that are associated with its use, and an informative annex placing the test in the context of the real electromagnetic environment to which it relates would help non-technical readers to assess its viability and applicability to any given case.
Elmac Services, Wareham, Dorset - Phone 01929 558279