PROF. CHRIS.C.

BISSELL

Models and black boxes : Mathematics as an enabling technology in the history of communications and control engineering / Modles et boites noires : Les mathmatiques comme technologie constitutive dans l'histoire des tlcommunications et de l'ingnierie de contrle
In: Revue d'histoire des sciences. 2004, Tome 57 n2. pp. 305-338.

Rsum RSUM. Depuis la fin du XIXe sicle, les ingnieurs des tlcommunications ont dvelopp un nouvel usage des mathmatiques au moyen des phasors , puis au moyen d'une modlisation de plus en plus labore des domaines temps et frquence. L'abstraction des composants systmes a conduit un mtalangage dans lequel la manipulation des configurations de circuits et d'autres reprsentations symboliques est devenue une consquence naturelle de la mathmatisation et progressivement un substitut. La systmique due aux ingnieurs des tlcommunications a t tendue d'autres domaines tels que l'automatique, et une synergie tout fait significative s'est instaure entre la modlisation mathmatique des systmes et des composants, la conception des instruments, le dveloppement du calculateur analogique et la conception des appareils et des systmes. L'histoire de cette approche ne parat pas avoir t tudie ni par les historiens des techniques, ni par les historiens des mathmatiques. Abstract SUMMARY. Front the end of the nineteenth century, communications engineers developed a new approach to the use of mathematics, initially through the use of phasors, and then through increasingly sophisticated time- and frequency-domain modelling. The abstraction of system components led to a metalanguage in which the manipulation of circuit configurations and other symbolic representations became a natural consequence of - and, increasingly, an alternative to - the mathematics. Systems ideas that originated with communications engineering were extended to other domains such as control engineering, and a highly significant synergetic relationship developed between the mathematical modelling of systems and components, the design of instrumentation, the development of the analogue computer, and the design of devices and systems. The history of this approach appears not to have been researched by historians of either technology or mathematics.

Citer ce document / Cite this document : BISSELL CHRIS.C. Models and black boxes : Mathematics as an enabling technology in the history of communications and control engineering / Modles et boites noires : Les mathmatiques comme technologie constitutive dans l'histoire des tlcommunications et de l'ingnierie de contrle. In: Revue d'histoire des sciences. 2004, Tome 57 n2. pp. 305-338. doi : 10.3406/rhs.2004.2215 http://www.persee.fr/web/revues/home/prescript/article/rhs_0151-4105_2004_num_57_2_2215

Models Mathematics in the

and

black

boxes :

as an history and

enabling control

technology engineering Chris Bissell (*)

of communications

RSUM. Depuis la fin du XIXe sicle, les ingnieurs des tlcommunicat ions ont dvelopp un nouvel usage des mathmatiques au moyen des phasors , puis au moyen d'une modlisation de plus en plus labore des domaines temps et frquence. L'abstraction des composants systmes a conduit un mtalangage dans lequel la manipulation des configurations de circuits et d'autres reprsentations symboliques est devenue une consquence naturelle de la mathmatisation et progressivement un substitut. La systmique due aux ingnieurs des tlcommunications a t tendue d'autres domaines tels que l'automatique, et une synergie tout fait significative s'est instaure entre la modlisation mathmat ique des systmes et des composants, la conception des instruments, le dveloppement du calculateur analogique et la conception des appareils et des sys tmes. L'histoire de cette approche ne parat pas avoir t tudie ni par les histo riens des techniques, ni par les historiens des mathmatiques. MOTS-CLS. Modlisation ; conception de circuit ; systmes linaires ; histoire des tlcommunications ; histoire du contrle. SUMMARY. Front the end of the nineteenth century, communications engi neers developed a new approach to the use of mathematics, initially through the use of phasors, and then through increasingly sophisticated time- and frequency-domain modelling. The abstraction of system components led to a metalanguage in which the manipulation of circuit configurations and other symbolic representations became a natural consequence of - and, increasingly, an alternative to - the mathematics. Systems ideas that originated with communications engineering were extended to other domains such as control engineering, and a highly significant synergetic rela tionship developed between the mathematical modelling of systems and components, the design of instrumentation, the development of the analogue computer, and the design of devices and systems. The history of this approach appears not to have been researched by historians of either technology or mathematics. KEYWORDS. Modelling ; circuit design ; linear systems ; history of tel ecommunications ; history of control. (*) Chris Bissell, The Open University, Milton Keynes MK7 6AA, Grande-Bretagne ; E-mail : cc.bissell@open.ac.uk. Rev. Hist. Sci, 2004, 57/2, 307-340

306

Chris Bissell

I. Introduction

Historians of telecommunications have tended to concentrate on enabling technologies such as the vacuum tube, the transistor, the microprocessor, and so on ; on the sociopolitical aspects of developing, regulating, and managing large-scale communication systems ; or on the social history of telecommunications (1). Yet just as significant historically - but largely neglected by historians of both technology and mathematics - is an approach to modelling, analysis and design based on a quintessentially com munications engineering use of mathematics. This approach, ultimately characterised by terms such as linear systems theory and black box analysis , is still a key factor in the develop mentof communications devices and systems, and highly influential in other areas of engineering. This paper addresses some aspects of the history of mathematical modelling in communications and control engineering, identifying a number of crucial features of such modelling, and setting them, albeit brie fly, in the wider context of communications and control enginee ring technology, including instrumentation, computation, and simulation. Consider, as a starting point, the following two quotations. 1 / In his general history of electrical and electronic engineering entitled From compass to computer W. A. Atherton writes : The mathematical treatment of circuits has passed through several stages of development. Originally mathematical statements were made describing the properties of individual components and groups of these components, which was sufficient for much 19th century work on tel egraphs and telephone networks. Later researchers analysed more complex circuit structures by breaking them down into smaller simpler sections for which mathematical statements could be made ; these sections were then used to analyze the whole complex structure. This stage began around 1900 [... and] out of this work emerged such disciplines as line theory, circuit theory and network analysis.

il) For an indication of the range of recent historiography see Bussey 1989 ; Fischer, 1994 ; Hugill, 1999 ; Huurderman, 2003 and Solymar, 1999.

Models and black boxes

307

haps the final stage, as exemplified by network synthesis, is the ability to synthesize a circuit by mathematically stating the function it is to have and then evolving the design that will achieve it (2). 2 / The second quotation is from a far more technical paper on the history of circuit theory, written by Vitold Belevitch, himself an important contributor to the discipline, to commemorate the fif tieth anniversary in 1962 of the ire, the (American) Institute of radio engineers. In Belevitch's view : Long before 1914 circuit theory had emerged, from general electr omagnetic theory, as an independent discipline with original concepts and methods. The conception of a circuit as a system of idealized lumped el ements is already firmly established - drawings of Leyden jars and rheost ats have gradually disappeared in favour of the now familiar graphical symbols. This assumes, at least implicitly, that a resistor is considered as a 2-terminal black box defined by the relation v = Ri, rather than as a phys ical device made of metal or carbon (3). Now, while both these quotations contain a good deal of truth, they also prompt us to exercise a little caution. The first presents us with a rather unproblematic view of the development of the use of mathematics in communications engineering - a somewhat Whig history to which, I suspect, most practising communications engineers would still subscribe. The second should sensitise us to the dangers of the temptation to ascribe to the past the viewpoint and assumptions of the present : the road to the view of a resistor as a 2-terminal black box , for example, was a long and complex one. This paper will try to tease out some of the historical context, and some of the processes that were involved in the broad histori cal development outlined by Atherton, as well as examine some influences of the approach beyond the confines of the control and communications communities.

(2) Atherton, 1984, 226. (3) Belevitch, 1962, 848.

308

Chris Bissell

II. Aspects of the development OF GENERAL LINEAR MODELLING THEORY

1 / Phasors A suitable point to start is with the application of complex numbers to electrical circuits, something that was promoted by a number of important figures in the last decade of the nineteenth century (including Arthur Edwin Kennelly and Oliver Heaviside). The idea was elaborated and popularised above all by Charles Pro teus Steinmetz, a German immigrant to New York in 1889, who soon became one of General Electric's most renowned engineers and inventors. In 1893, Steinmetz gave a lengthy address to the International Electrical Congress in Chicago. In this talk he clearly stated the advantages of the vector (phasor) approach (4) : The method of calculation is considerably simplified. Whereas before we had to deal with periodic functions of an independent variable time , now we obtain a solution through the simple addition, subtrac tion, etc., of constant numbers [...] Neither are we restricted to sine waves, since we can construct a general periodic function out of its sine wave components [...] With the aid of Ohm's Law in its complex form any cir cuit or network of circuits can be analysed in the same way, and just as easily, as for direct current, provided only that all the variables are all owed to take on complex values (5). Incidentally, of this paper, Steinmetz later wrote : ... there was no money to publish the Congress paper, and the paper remained unpublished for years, and the symbolic method unknown (6). Yet the paper did appear in German the same year as the Cong ress, and it is from this source that the above citation is taken. Although the use of complex numbers appears to have mysti fied many early electrical and electronics engineers - Steinmetz's (4) In the phasor approach a sinusoid is characterised by a complex number (phasor) V = A exp (i ), such that |V| = A represents the amplitude of the sinusoid and Arg V = represents the relative phase. An ideal linear electrical component can thus be characterised by a complex gain G, so that |G| represents the ratio of the amplitude of the output sinusoid to that of the input, and Arg G represents the phase shift introduced by the component. (5) Steinmetz, 1893, 597, 631. (6) Anon, 1996, 11.

Models and black boxes

309

Chicago talk is stated to have been met with much incompreh ension,for example - the adoption of the phasor approach was highly significant for future developments in two particular ways. First, as noted by Steinmetz himself in the previous quotation, it allowed quite complicated calculations in the time domain to be replaced by much simpler ones in terms of frequency. Second, it was indeed an important step towards the black box concept. The defining equations for resistors, capacitors and inductors were all subsumed into a generalised, complex version of Ohm's relationship ; and even if it would be premature to talk of implicit 2-terminal black boxes at this time, such a representa tion of components as complex impedances was clearly a great conceptual step. Furthermore, much greater attention in the manufacture of such devices was subsequently paid to improving the closeness of the approximation of the component behaviour to the ideal mathem atical model. A great deal of engineering effort was devoted to ensuring that resistors, capacitors and inductors behaved as linearly as possible over a wide range of operating conditions. Indeed, in this way the mathematical model became a specification for the manufacture of devices, rather than an analysis of their behaviour. An analogy will perhaps make this clearer. A geometric al progression can be used to model inflation in an economy ; this is an analysis of more complex behaviour that can be modelled in this way. But if a bank offers compound interest on a deposit, this is a specification of how the interest will accrue, based on the same mathematical model. In terms of the simplest electrical device, the resistor, an analysis of various materials shows that Ohm's rela tionship holds roughly over a given range of conditions. Using Ohm's relationship as a specification, however, means that manuf acturers have to find ways of avoiding as far as possible deviations from non-linear behaviour. Typically, therefore, materials are cho sen - or developed specifically, such as manganin at the end of the 19th century - whose resistivity varies as little as possible with tem perature ; and the physical construction of the device is designed so as to dissipate heat generated into the environment. Or, when using a wire-wound resistor, care is taken to ensure that inductive effects are negligible under normal operating conditions.

310

Chris Bissell

2 1 Filter design and circuit theory In the years following the first publications on the subject by Steinmetz and Kennelly, phasor analysis became a basic tool of all electrical, electronics, and communications engineers. During the first two decades of the twentieth century communications engi neers made enormous progress in the use of time- and frequencydomain techniques, and Heaviside's operational calculus was put on a much firmer mathematical footing by George Ashley Campb ell, John R. Carson, and others. Most of these figures were asso ciated with Bell Labs, one of the most important centres for research and development in telecommunications from the late nineteenth century onwards (7). Three closely related approaches were of great significance : 1 / the operational calculus itself, which associated mathematical variables in a particularly direct manner with physical quantit ies, and greatly assisted engineers in the manipulation of such quantities as part of analysis ; 2 / the mathematical technique of convolution (about which a little more later) used to calculate the input-output behaviour of linear systems in terms of time-varying waveforms rather than frequency responses ; 3 / the use and dissemination of Fourier analysis to calculate inputoutput behaviour in the so-called frequency domain . A driving force for technological development at the time was the need to exploit bandwidth effectively for both carrier telegra phy and the newer telephony. The solution was so-called frequen cy-division multiplexing (fdm) (8), and for successful fdm, wave filters with quite stringent pass-band characteristics were needed in order to select the desired channel without excessive distortion. The wave filter was invented independently by Campbell in the usa in 1909 and Karl W. Wagner in Germany a few years later. In developing design techniques for such filters, communications engi(7) See Fagen, 1975 and O'Neill, 1985, for a general history. (8) In FDM, different channels are allocated different parts of the overall bandwidth of a transmission medium. The best-known example is in analogue radio and television broadc asting, where different channels are allocated to different parts of the radio spectrum.

Models and black boxes

311

neers started to use mathematics in a radical new way. In particul ar, circuit diagrams became what we might term a metalan guage for the mathematics. Readers are referred in particular to the classic 1920s papers by George Ashley Campbell, Otto Zobel, Ronald Foster and Wilhelm Cauer (9). Complex filters were repre sented as a series of interconnected sections, then elaborated as a set of equivalent circuit configurations. What is beginning to emerge at this time is the distancing of circuit diagrams in the design phase from their eventual implementation : the manipulat ion of circuit component symbols (in general, complex impedances) becomes an attractive alternative to the manipulation of mathemat ical symbols. Ultimately this will lead to the use of prototype cir cuits in filter design that bear little or no resemblance to the ult imate electronic circuit.

3 / Duality and black boxing I do not wish to give the impression that this approach explo ded fully-formed on to the filter design scene in the early 1920s ! The work of Campbell, Zobel and others drew on a great deal of earlier work. One of the most important concepts, developed towards the end of the 19th century, was that of duality. As engi neers and scientists elaborated the theory of electrical circuits, they noticed a number of parallels between the variables and approa ches they used. As a simple example, consider the defining equat ions for an ideal capacitor and inductor in terms of the relation ship between voltage V and current I, where is the capacitance and L the inductance, dt dt

respectively. These expressions are equivalent under the transfor mations VI and L <=> C. Relationships such as this allowed engineers to draw up a list of corresponding variables and circuit concepts, such that representations in terms of one could easily be (9) Campbell, 1922 ; Zobel, 1923, 1924; Foster, 1924, and Cauer, 1926. The first three researchers were at Bell Labs ; Cauer was a student at the Technische Hochschule, Berlin, and then employed at Mix & Genest, a German Bell subsidiary.

312

Chris Bissell

converted into an equivalent representation in terms of its dual. Examples include : voltage resistance inductance short-circuit series node current conductance capacitance open-circuit parallel mesh

For example, as early as 1899 Kennelly had established the fun damental star-delta transformation where three devices in a star configuration were shown to be equivalent to three other devices in a triangular configuration, providing certain relationships, based on the concept of duality, held between the individual devices in the two configurations. Such abstraction away from the characteristics of individual devices towards a system-theoretic approach soon made itself felt in practical design techniques. Duality allowed alternative, equiva lent circuit configurations to be identified, and by the early 1920s wave filter design was already pointing firmly towards the notions of black box and linear systems that were to become so cru cial over the following decades. The concept of the linear black box appears to have been made explicit for the first time by Zobel (10). In this approach, a complex linear electrical network is represented by its input-output behaviour ; at this level of abstrac tion the precise nature of the interconnections of components inside the black box becomes irrelevant. A modern view of such black-boxing is shown in figure 1 (11), with particular reference to the relationship between time- and frequency-domain inputoutput relationships. (Note especially the equivalence of convolu tion in the time domain and multiplication in the frequency domain.) (10) Zobel, 1924, 572. (11) Note that this modern view, included here to aid discussion, is significantly diffe rent from Zobel's general linear transducer model of 1924, although the general prin ciples are identical.

Models and black boxes linear system

313

Fourier transform transfer function I frequency response

Y(j))=X(Jc)xH(Jc) Fig. 1 . and A linear time by black input-output box defined relations. in frequency

4 / Circuits and metalanguages The mathematicization of filter design - and the correspon ding further development of the electrical metalanguage - conti nued apace during the 1920s (12). As a later commentator put it : Foster partitioned the given rational function into a sum of partial fractions that could be identified easily as a series connection of impedanc es or a parallel connection of admittances [duality again]. Wilhelm Cauer expanded the rational function into a continued fraction representing a ladder network. Each method gave two alternative networks, which were called canonical forms because they could always be obtained from a rea lizable immittance function and because they employed a minimum numb erof elements (13). Cauer's 1926 paper was perhaps the culmination of this stage in the development of filter synthesis, and was a shortened version of his doctoral thesis. He had been in contact with Foster in the mid 1920s, and retained strong links with researchers in the us, spending a year at mit in 1930 (14). He and his contemporaries showed how quite sophisticated mathematics (partial fractions and continued fractions of a complex variable) specifying desired filter behaviour could be translated directly into circuit configurations, in a way that allowed the circuit designer to move at will between (12) Foster, 1924, and Cauer, 1926, are two classic papers. (13) Zverov, 1966, 130. (14) Cauer et al., 2000, includes an extensive technical review of his contributions.

314

Chris Bissell

the two representations. Increasingly, designers began to think of filter specifications in terms of these so-called prototype circuits that corresponded to the mathematical model, rather than in terms of the mathematical representation itself. It is important to unders tand that such prototype configurations, even though expressed as networks of inductors and capacitors, are not necessarily the way that the filter itself is finally realised : rather, such prototypes became increasingly a design tool expressed in a language that communications engineers could understand without reference to the mathematical theory they embodied. The fact that this subse quently became common currency in electronics design should not blind us to the novelty of the approach. This approach went much further in subsequent decades, with considerable blurring between real circuit components and mathe matical abstractions. Writing in an editorial in the September 1955 edition of the ieee Transactions on circuits and systems, the editor, W. H. Huggins, wrote : ... modern circuit theory is concerned but little with the circuit as a physical entity and, instead, has become increasingly involved with [...] signals [...] Thus the usual circuit diagram may be regarded as a pictorial form of a signal flow graph [i.e. an alternate mathematical representat ion] [...] (15). As the years went by, electronics engineers introduced a whole set of new, ideal, circuit elements - such as the gyrator, nullator, norator, or supercapacitor - which became just as real to the circuit designers as resistors, capacitors and operational amplifiers. Some of these devices are non-realisable, yet can be exploited in a highly effective way at the design stage. They are real in the sense that they lead to useful ways of talking about the cons traints within a network without writing down the corresponding mathematics per se. This particular pictorial representation of circuit constraints is very powerful and is as formal as a convent ionalmathematical description. It is a way of writing the mathe matics in terms of connections of components. Moving backwards and forwards between these mathematical and circuit-symbolic domains corresponds to doing very complex mathematics simply by modifying a circuit diagram or using a network transforma(15) Cited in Huggins, 1977, 667.

Models and black boxes

315

tion. Furthermore, the approach has the advantage of delivering to the designer a very good intuitive idea of what the result of design changes will be.

III. Systems thinking

A second major argument of this paper is that the approach to mathematics developed initially by communications engineers led to a much wider applicability in technological analysis, modelling and design. The work presented so far, however, does seem to have been directed only to the narrower domain of electronics design, particularly for telegraphy and telephony. But some engineers in the 1920s were already making the intellectual leap beyond their own discipline. The German Karl Kupfmuller, for example, an engineer at Siemens & Halske ag, Berlin made the point as early as 1928 : Practical problems are generally concerned with the relationship bet ween two system variables ; for example, between a force Sx and a dependent variable S2. Assuming a linear system, then in the steady state : S2 = U.S, [...] where U is a factor independent of S,. Using symbolic vector [phasor] notation, U is in general complex, and depends on frequency : U = A(co) exp[- 1 a(()]. We shall call A(o>) the transmission factor [gain] and () the trans mission angle [phase shift] of the system. The variables Sj and S2 can be completely arbitrary in nature ; St for example could represent a mechanical force, an electrical voltage, or a sound pressure level ; S2 a velocity, a magnetic field, or the strength of an electric current. [...] The complex transmission factor provides a complete description of the system with respect to the variables S, and S2 [and...] determines the form of the transient behaviour (16). This appears to be the first explicit suggestion that the approach developed for electrical systems could also be applied to other technological devices. What is more, Kupfmuller had also (16) Kupfmuller, 1928 b, 19-20.

316

Chris Bissell

been looking beyond the confines of deterministic models. In a 1924 paper he analysed filters characterised by the ideal cha racteristic of constant amplitude ratio and linear phase. Such a fi lter characteristic, while linear, is unrealisable and non-determ inistic,yet can be used to derive extremely useful practical rules of thumb for system design. For example, Kupfmuller showed that his ideal frequency domain constraints could be translated into corresponding time-domain constraints on rise time and transient response. In other words, the precise way in which the filter was implemented was, to some extent immaterial : any bandlimiting brought with it an inescapable restriction on time-domain beha viour (17). It is another example of duality : the stricter the bandli miting, the greater the rise time - something that was generalised as the notion of time-bandwidth product. Kupfmuller's approach was not immediately accepted by the profession, as he explained fifty years later : There were serious objections to my 1924 paper on transient re sponse in wave filters. At that time it was usual to calculate transient re sponse only for concrete devices such as cables, transmission lines, net works, etc. That it was possible to make abstract assumptions about the characteristics of systems, and thereby, particularly in complicated cases, arrive simply at approximate conclusions about transient response, led to many objections being raised. Moreover, neither did the phrase systems theory for such an approach find immediate acceptance (18). By the mid 1930s, however, general linear systems theory (at least for deterministic signals) was firmly established within commun ications engineering, and the first textbooks to deal with the topic had appeared. One classic is Communication networks published in two volumes in 1931 and 1935 by Ernst Guillemin, who subsequent ly proved to be a prolific writer of highly influential student texts (and a great engineering educator at mit). In the preface to the 1935 volume he pointed out the wider applicability of the subject matter : The communications viewpoint is chosen for the more detailed dis cussions because it affords at present a more suitable vehicle for illustra ting the fundamental principles. [...] Nevertheless these ideas and prin ciples are too general in nature to remain confined to one field of application, and are more recently beginning to make their appearance in the treatment also of power problems (19). (17) Kupfmuller, 1924, 146. (18) Kupfmuller, 1977, 775. (19) Guillemin, 1935, III.

Models and black boxes

317

Guillemin was well-acquainted with the German literature on filter design, having studied in Munich under Arnold Sommerfeld and he incorporated Wilhelm Cauer's approach to filter synthesis as well as Kupfmuller's general linear theory into this volume. A new type of language is also reflected in the book, in which the mathematics starts to become remarkably concrete. For example, when discussing Foster's reactance theorem, Guillemin writes : ... a driving-point reactance function is uniquely specified by the location of its internal zeros and poles (20) plus one additional piece of information which [...] takes the form of a multiplying factor [...] (21). In this new language the poles and zeros of the filter become just as real as the resistors, capacitors and inductors from which it is constructed. We shall return to this point later.

IV. Interlude : engineers, mathematics and

systems

At this point it is worth spending a little time on a closer exami nation of the rle that mathematics plays in engineering, particul arly (but by no means exclusively) in communications and control engineering. Engineers use a variety of models to understand the systems or devices that they design and build. Mathematical models - like other models such as graphical representations or scale models - are used for many different purposes. They can be used to provide a basis for explanations of observed behaviour, for example, or to predict the behaviour of existing or non-existing systems under particular conditions, or to design new artefacts that will behave in a prescribed way. Perhaps most importantly, models are starting points for conversations among practitioners about the systems they are claimed to represent. Models have to be mediated and negotiated within a community of practice to make any sense. As part of their development, engineers learn how to talk about (20) Poles and zeros correspond to the roots of the rational frequency-domain transfer function of a linear system, as discussed in more detail below. (21) Guillemin, 1935, 186.

318

Chris Bissell

their models : they learn what stories to tell about them and to recognise what sorts of conversations are legitimate (22). For this reason, the activities and conversations relating to a particular mathematical model within an engineering environment are very different from the activities and conversations relating to the equivalent model in a mathematical environment. Take the ear lier discussion about duality, for example. A mathematician might be primarily interested in the solution of the differential equation model of an inductor or capacitor (or a more complex circuit) for arbitrary forcing functions. The electronics engineer, however, is likely to be much more interested in generic behaviour, such as the response to an idealised forcing function such as a step input, or the very abstraction that gives rise to the concept of duality. Focus ingtoo closely on the mathematical details of a particular model may not be helpful when the goal is to understand circuit or system behaviour better. It should not be surprising, therefore, that what counts as a valid engineering approach often differs from what counts as a valid mathematical approach. It will be useful to bear these remarks in mind in what follows below.

V. Feedback systems and control engineering

One of the most influential applications of the communications engineering way of doing mathematics was to feedback systems : first to the feedback electronic amplifier, and later to other feedback control systems. The classic papers are Karl Kupfmuller (1928 b), Harry Nyquist (1932), and Harold Black (1934) (23). Black conceived the negative feedback amplifier in 1927, by which time the notion of electrical/electronic black boxes was well established, although the work was not published until 1934. The major impetus for the work was the need to develop amplifiers with stable gain for use in transcontinental telephony. Existing amplifiers fluctuated too much in gain to be used without daily (22) For a more extensive discussion see Bissell and Dillon, 2000, 6. (23) Kupfmuller has already been mentioned in section 3 above. Harry Nyquist and Harold Black were researchers at Bell Labs.

Models and black boxes

319

manual adjustment, and noise and distortion were also problemati c. Black exploited feedback to linearize a high-gain amplifier and to render it comparatively immune to noise and disturbances. As he wrote : ... by building an amplifier whose gain is deliberately made, say 40 decibels higher than necessary (10,000 fold excess on energy basis), and then feeding the output back on the input in such a way as to throw away the excess gain, it has been found possible to effect extraordinary improvement in constancy of amplification and freedom from nonlinearity (24). Black's analysis, illustrated in figure 2, is simplicity itself, and a telling example of the power of the systems and black-box approach. e 3 (E+N+D) feedback circuit P Fig. 2. Black's analysis of a feedback amplifier. The amplifier, with high (complex) gain (x, amplifies the sum of the two inputs (e plus the feedback signal), and introduces noise n and distortion d. The feedback circuit, a passive network with gain p < 1, returns a proportion of the output to the input. In Black's terminology, the total output consists of the amplified wanted signal, E, together with the overall noise N and distortion D (the latter two terms including the effect of the feedback), such that : E+N +D

and hence 1 (24) Black, 1934, 1-2. 1 1 -

320

Chris Bissell

If [(3 > 1, then both the noise N and the distortion D can be made very small, and the wanted output E is simply - el$ - that is, it depends only on the characteristics of the feedback circuit, and is independent of the (fluctuating) gain of the amplifier. Note the high level of abstraction in figure 2, far from the actual circuit design ultimately used to implement the feedback amplifier by means of vacuum tubes and other electronic components (25). From a mathematical point of view the analysis is naive and non rigorous : it is a purely static analysis and fails to consider either the dynamics of signal behaviour, or the conditions under which the feedback system remains stable (26). Yet from the engineering point of view it enables the relevant questions about noise, distortion and constancy of gain to be posed ; and, indeed, it enabled functional long-distance telephony to be implemented (27). By the time Black and Nyquist analysed electronic feedback cir cuits in the frequency domain, their German contemporary Kupfmuller (1928) had already carried out related work in the time domain (prompted by the analysis of an automatic gain control cir cuit, but developed generically) - including the statement of a valid stability criterion. Figure 3 shows his generic loop consisting of separate black boxes to represent the system to be controlled and the control mechanism generating the necessary control action. He opened the loop - as indicated in part (b) of the figure, and as did Nyquist a few years later - in order to analyse loop stabi lity(28). The adoption, by engineers concerned with control problems, of the analytical and design techniques of communications enginee ring and filter design was by no means straightforward and unproblematic. General systems thinking of the type exhibited by Kupfmuller appears to have emerged independently in the usa, uk and Germany in the late 1930s and during the war. The us war effort in this field was characterised by intensive activity at centres such as Bell Labs, the radiation laboratory and the servomechanisms laboratory at mit, and elsewhere - work coordinated effecti vely by the National Defense Research Committee and the Office (25) Black's original 1927 sketch of his idea, made on a copy of the New York Times, can be found at http://www.wpi.edu/Academics/Library/Archives/Pubs/ECE/Images/16.gif. (26) A problem fully analysed by Nyquist in 1932. (27) For further discussion see Mindell, 2000, and Bennett, 1993. (28) Kupfmiiller's work is discussed in greater detail in Bissell, 1986.

Models and black boxes

321

(a)

Fig. 3. Kiipfinilller's analysis (b) of a generic feedback system.

of scientific research and development directed by Vannevar Bush (who reported directly to the president). In the the less formal body known as the Servo-Panel played an analogous rle, faci litating communication between the various industrial and govern ment research centres working on control problems. In both count ries the key to success was the assembly of interdisciplinary teams in which the novel approaches developed by communications engi neers could be transferred to new areas (29). The wartime situation in Germany was much more fluid, with little national coordination in the field of control engineering. One important body, however, was the control engineering terminology committee of the Verein Deutscher Ingenieure (vdi), through which engineers from different backgrounds were enabled to come together to improve standardi zation of the new discipline. (29) The organisational and technical aspects of this wartime environment have been researched in detail by Bennett, 1993, and Mindell, 2002.

322

Chris Bissell

Engineers from various countries have reflected on the difficulty of applying the new ideas and moving from a communications way of thinking to a control or systems way of thinking. For example, the Briton Arnold Tustin (an electrical engineer by training), who worked on the fire control (30) problem recalled half a century later : I think my personal realisation [of the systems approach] came as I worked with control systems which consisted of a number of operations in sequence - the fire control problem was classic, of course, with target detection, tracking, prediction, and gun laying. As we worked on such systems, combining the responses of the various elements to get the total response, it became natural to think in general system terms, applying identical methods even though one element might be electrical, another mechanical, hydraulic or even human (31). Similarly, Winfried Oppelt, a German engineer (initially an applied physics graduate) who worked primarily on flight control systems later noted : Systems thinking emerged very slowly, and really in a quite anony mous way - no single person is responsible, even though some indivi duals, such as Kupfmuller, were more alive to the concept at an early stage than others. We also devoted a lot of effort to this aspect in the vdi [German engineers' professional body] Specialist Committee. But the development of systems thinking came about very gradually, as part of group work and discussion. It was a long time before the general applicab ility of control concepts was understood. For example, F. Strecker, at Siemens, had come up with a version of the Nyquist criterion in 1930, although it wasn't published at that time, and most people, including myself, were not aware of the work at all until much later. After the war he incorporated his original paper into a book, but if you read it you can see that from the general control point of view he had not taken the deci sive step, at least not in my opinion. The shift from a communications way of thinking to a much more general control way of thinking was the fundamental step, and that had still to be made (32). And the American control engineer, Ed Smith, speaking at a time when the new techniques were just emerging from wartime secrecy, went so far as to say : The experts in the servomechanism and control fields are unfortunat ely not exceptions to the rule that the experts in an art always ball up the (30) Fire control was the system for directing and firing servo-controlled anti-aircraft guns. (31) Bissell, 1992 b, 226. (32) Bissell, 1992 a, 20.

Models and black boxes

323

terminology and notation until they can be followed only by a like expert [...] At least the [process engineer] can directly express a ratio of effect to cause [...] without becoming involved in the width of a hypothetic al frequency transmission band and in such acoustical abstraction as dB per octave (33). The exasperation of process control engineers trying to come to terms with the new language of communications engineering and filter design is only too evident ! In particular, the accounts of the discussions at conferences and learned society presentations testify to the rather difficult process in adopting this systems way of thinking, and the novelty of applying general systems ideas outside their natural home of communications and electronics engineering. It was not a question of simply applying existing communications theory and techniques to other systems. Indeed, the existing theory was often not well known to the engineers pioneering the development of the new clo sed-loop systems. Rather, the collaborators from different technical backgrounds renegotiated their modelling techniques, and the lan guage in which they were expressed, so as to abstract the essential. The detailed history of this seminal shift in thinking calls for more research, but certainly includes the following key features : the interaction between theory and practice, rather like the inte raction between mathematics and circuits in the context of filter design ; the importance of the social dimension, such as interdisciplinary teamworking, standardization bodies, government initiatives, and learned society discussions in facilitating the crucial diss emination of linear systems theory within the wider engineering community. These influences were particularly noticeable in the us, UK and Germany (34).

(33) Discussion of Brown and Hall, 1946, 523. (34) Further discussion of the importance of the new language of control is to be found in Bissell, 1994.

324

Chris Bissell

VI. Models and reality In the 1920s and 1930s the modelling approach developed by communications engineers led to considerable unease about quite how the mathematics related to the real world . For some, the frequency-domain models were particularly problematic. John Bray, a senior engineer in the British Post Office and subsequently British Telecom notes in his book on the history of telecommunicat ions that : It seems remarkable now that in the 1920s there were some, inclu ding the eminent scientist Sir Ambrose Fleming (35), who doubted the objective existence of the sidebands of a modulated carrier wave, regar ding them as a convenient mathematical fiction [...] (36). Fleming had sparked off a vigorous debate in the pages of the journal Nature in 1930, when he called into question the practical significance of the sidebands in the mathematical description of an amplitude-modulated carrier wave. When a sinusoidal carrier is amplitude-modulated by a message signal, the composite spectrum consists of the carrier plus the spectrum of the modulated signal reflected about it, as shown in figure 4. Yet Fleming wrote : The carrier wave of one single constant frequency suffers a variation in amplitude according to a certain regular or irregular law. There are no multiple wavelengths or wave bands at all (37). Fleming was certainly well aware of Fourier theory : The complex modulation of a single frequency carrier wave might be imitated by the emission of a whole spectrum or multitude of simulta neouscarrier waves of frequencies ranging between the limits n + N and n - N, where n is the fundamental carrier frequency and N is the max imum acoustic frequency and 2N is the width of the wave band. This, however, is a purely mathematical analysis, and this band of multiple fre quencies does not exist [...] (38). (35) Fleming was a pioneer in the development of electric lighting and English tel ephony/telegraphy, and the inventor of the thermionic valve. He taught at a number of English universities, becoming a professor at University College London in the UK's first department of electrical engineering. (36) Bray, 1995, 71-72. (37) Fleming, 1930, 92. (38) Ibid. (My italics.)

Models and black boxes

325

The debate in Nature continued for several months. John Bray returns to this in the conclusion to his book, in a personal reminiscence : In 1935 [...] I entered the Open Competition for a post as Assistant Engineer in the Post Office engineering department. This included an interview with the Chief Engineer, Sir Archibald Gill - an interview during which he asked : Do you believe in the objective existence of side bands ? It may seem remarkable that such a question should even have been posed, but at the time there was still a lingering controversy [...] as to whether sidebands were just a convenient mathematical fiction or whether they were real . Luckily I had witnessed [...] a convincing demonstration involving a frequency-swept tuned circuit and a sine-wave modulated carrier, the response being displayed on a primitive electr omechanical Dudell oscilloscope and revealed as a triple-peaked curve. So my response to the question was a triumphant Yes, I have seen them (39) ! This is an extraordinarily revealing anecdote. To Bray in 1935, the spectrum of the modulated signal displayed on the oscilloscope was just as real as the corresponding time-varying waveform. In fact, both types of display are fairly remote from reality : like a spectrum, an electrical voltage or current can only be revealed by an instrument designed to detect it. The fact that the time-domain representation of such variables often seemed (and still seems) so much more natural has more to do with three centuries of natural science and its particular models and conventions than to any ontological distinction (40). Communications engineers developed their instruments and practical techniques hand in hand with their mathematical models and other symbolic representations. In a sense, of course, this was not so different from the experience of natural scientists ; what was different, however, was the way the modelling and instrument ation techniques so developed were soon exploited for design. So, for example, the frequency-domain models of modulated wave forms led to a whole range of practical techniques which derived directly from the mathematical models. Single-sideband, sup pressed carrier, amplitude modulation originally suggested by (39) Bray, 1995, 356. (40) Although it must be admitted that our natural models of light and sound tend to be in terms of spectral notions (colour, pitch).

326

Chris Bissell

John Carson in 1915, is one of the most impressive examples of this approach. Carson proposed that, since all the information of the message signal in figure 4 is contained in each sideband, one of the sidebands as well as the carrier can be suppressed, thus saving bandwidth. The sidebands of a number of message signals can then be combined to use a single channel by means of the technique of frequency-division multiplexing mentioned briefly above. The technique is illustrated in figure 5. Each sideband is translated in frequency by the appropriate amount before trans mission, and then recovered by the reverse process on reception. It is particularly interesting to note the language used by commun ications engineers in describing this process. The sidebands are treated very much as physical objects : they are shifted and combined so as to fill the available bandwidth. They have become something very different from the mathematical spectra (Fourier transforms) of the message signals.

spectral density carrier upper sideband frequency Fig. 4. The spectrum of an amplitude-modulated signal.

message message message message

1 2 3 4

message message message message

1 2 3 4

frequency Fig. 5. The principle of frequency-division multiplexing.

Models and black boxes

327

VII. Some post-war developments

The period following the Second World War was characterised by a number of important developments in what might be called the epistemology of engineering modelling, directly influenced by communications and control engineering. Three examples will be considered very briefly here : the rise of general-purpose analogue computing ; the realisation of new instruments and devices ; and aspects of the modern exploitation of the digital computer.

1 / Analogue computers Analogue computing devices - in other words, devices in which physical variables are manipulated directly as analogues of mathem atical variables in order to compute solutions - have a long his tory. The slide rule is an early example, and various highly sophis ticated mechanical and electro-mechanical devices were developed in the 19th and early 20th centuries ; the prewar differential analy zers designed to solve the ordinary and partial differential equat ions of dynamics and fluid mechanics are perhaps the best-known of the latter (41). In the late 1940s and 1950s, however, a new gene ration of general-purpose analogue computers appeared, influenced strongly by the technologies and modelling approaches outlined above. The crucial insight, which had occurred simultaneously in a number of countries during or just after the war, was that the tech nologies used for high-performance communications and control systems - in essence, the direct descendants of Black's feedback amplifier - could be used to solve a wide range of mathematical problems. In particular, since electronic networks could be realised (41) For a discussion of work on differential analyzers by Vannevar Bush and coworkers see Small, 2001, 40-45, and Bennett, 1993, 97-104. Norbert Wiener was another important pioneer in this and related fields ; for a brief description and further references see Bennett, 1993, 104-105 and 176-181. For a consideration of the interaction between mathem atics and technology in the design and development of these early computational devices, see Puchta, 1996.

328

Chris Bissell

which implemented quite accurately the mathematical operations of integration and differentiation, suitable interconnections of such networks could be used to solve differential equations directly. And such analogue computers often gave engineers a feel for the problem under consideration very akin to the feel of filter desi gners for the behaviour of their prototype designs discussed above. An extended metalanguage was created in which system behaviour was expressed directly in terms of the components of the analogue computer (42). As the author of an early textbook on analogue computing remarked : in contrast to the mathematician, who might analyse a system, write down the differential equations, and then solve them, ... the engineer would examine the system and would then build a model of it, which he would call a simulator. He would then obtain solu tions of his problem, perhaps without even having written down the full set of equations [...] (43). Analogue computers coexisted with digital computers for seve raldecades, even though they were ultimately eclipsed by the lat ter. In the early days, analogue computers offered many advanta ges for engineering modelling, particularly in the fields of aerospace and missile control. For example, what they lacked in precision, they gained in speed. A typical application of the immed iate postwar Project Cyclone , funded by the us Navy, was the simulation of a guided missile in three dimensions. The ave rage runtime for a single solution on the analogue computer faci lity was approximately one minute, while the check solution by numerical methods on an ibm cpc (card programmed calculator) took 75 hours to run (44). The epistemological aspects were just as significant. James Small notes : ... the use of electronic analogue computers led not only to the growth of tacit forms of technological knowledge but also to the revision of engineering theory. They enabled engineers to build active models that embodied and operationalised the mathematical symbolism of engineering theory. With these models, empirical methods could be used to study the behaviour of a technological system beyond the limits predicted by engi(42) For a history of analogue computing see Small, 2001. (43) Wass, 1955, 5. (44) Small, 1993, 11.

Models and black boxes

329

neering theory. Electronic analogue computers helped designers bridge the gap between the limits of earlier empirical methods, current analytical techniques and the real-world systems they were constructing (45).

2 / New instruments and devices A closely related phenomenon is the way that new, specialpurpose devices were produced to implement particularly import ant mathematical operations. For example, during the Second World War, linear systems theory had made enormous strides, par ticularly in the handling of random signals and noise (for gun aiming predictors, for example). One particular technique, known as correlation, proved to be extremely fruitful in the theory of signal detection. The correlation function can be written :

The process of correlation thus involves taking two wave forms x(t) and y(t), and comparing their similarity for varying di splacements T in time against one another. Around 1950, engineers at mit built an electronic device to carry out this operation and plot the correlation function on a chart recorder. Since then, cor relation functions have become just as much a part of the lan guage of communications engineers as spectra, and are implement ed in a huge range of modern electronic devices (digital processors proved to be highly amenable to such tasks). Such inte raction and synergy between mathematical models, metarepresentations and electronic instrumentation was a decisive feature of the development of communications engineering and enginee ring modelling in general. In their conclusion to their paper reporting the mit electronic correlator, the authors remarked : The method and technique of detecting a periodic wave in random noise presented here may be regarded as a type of filtering in the time domain [...]. From an engineering point of view, many equivalent opera tions may be more practicable and feasible in the time domain. For ins tance, a zero bandwidth filter in the frequency domain corresponds to (45) Small, 2001, 274.

330

Chris Bissell

an extremely stable oscillator in the time domain. At the present time it is much easier to build a stable oscillator than to build a network having zero bandwidth (46). Yet again we see the modelling flexibility of moving between time- and frequency-domains, and the way in which the mathemat ics has become something quite different in the hands of the com munication engineers. And, most interestingly of all, perhaps, the mathematical black box entitled a correlator has become a physical black box that delivers ink traces on paper of correlation functions !

3 / Engineering modelling and the digital computer From the 1950s onwards, the availability of digital computation became increasingly important for the design of engineered systems and their components (47), as well as for general-purpose enginee ring computation. For a long time, the use of digital computers for engineering applications implied a predominantly mathematical approach - in particular the design of robust numerical algorithms and their use for solving differential equations. Over the last decade or two, however, vastly increased computing power, the development of graphical user interfaces, the introduction of symbolic computation, and the ubiquity of the pc have led to a significant change, with a remarkable convergence of design tools, simulation, and implement ation. It is particularly ironic, perhaps, that the digital computer, having effectively vanquished its analogue counterpart, has now much in common with the latter in so far as the human-computer interface is concerned. Whereas earlier engineers would physically interconnect operational amplifiers and other devices so as to simul ate a problem, the modern designer is likely to drag symbols representing the various operations into position on screen, and interconnect them as appropriate, before running a numerical or symbolic simulation or analysis. What the two approaches have in (46) Lee et al., 1950, 1171. (47) Zverov, 1966, points out, for example, that the later filter design methods of Cauer and Sidney Darlington, dating from the 1930s, only became popular with the advent of cheap computation.

Models and black boxes

331

common is that modern digital computer tools now employ the same metalanguages as the engineer, and the user can again carry out highly sophisticated modelling or simulation activities without employing very much mathematics per se.

VIII. A NEW ONTOLOGY?

In this paper I have tried to tease out some of the particularities of the communications and control engineering approach to mathematics and modelling, in an attempt to understand how this approach evolved. In one sense, the mathematics used is highly abstract and complex indeed (Fourier transforms, correlation func tions, complex variable theory, etc.). In contrast, though, many of the manipulations carried out using the engineering models are remarkably simple and non-mathematical . As we have seen, such models have led to engineers talking a language very different from that of conventional mathematics. This has turned out to be a remarkably concrete language. We have already seen this in the context of the manipulations of circuit components or the spectra of signals. As a final example, consider the pole-zero approach, much used by control system and filter designers. For readers unfamiliar with complex variable theory, the poles are the values of a variable that render a complex func tion infinitely great, and the zeros are those that render it zero. A linear system can be modelled by a transfer function H(s) where s is the Laplace variable. In many common systems H(s) is a rational function such that : ft(s) = K(J + *i)Cy + z2)"-( S + Zm) ft(s) K (s + p{)(s + p2)...(s + pn) Here, z and p are (in general complex) values known as the poles and zeros respectively, and is a real constant known as the system gain. When engineers use such models they move away completely from the mathematical language of complex variable theory. They describe systems as possessing so many poles and zeros ; they

332

Chris Bissell

talk about the need to place poles or zeros in certain regions or to move them to a more favourable position ; and they become highly adept at visualising the pattern of poles and zeros on the complex 5-plane in terms of how they want the system to behave. Figure 6 shows a very simple example, a first-order system with no zeros and one pole (indicated by the small cross). The fr equency response of the system can be visualised in terms of a cut through the |H(j)| three-dimensional surface along the imaginary axis, and engineers develop a remarkable ability to visualise how this surface changes as the pole moves around. 3-D surface for |10/(s+10)| s-plane amplitude frequency response curve

-10

Fig. 6. Modelling a simple linear system in the s-plane.

This is an example of electronics or control engineers doing mathematics . The important point is that the language they use is not the traditional language of mathematics, even if the manipulat ion of their models may be completely analogous to other manipul ations of other, more conventionally mathematical models (and often even isomorphic in a strict mathematical sense). Moreo ver, this linguistic shift is more than just jargon, and more than just a handy way of coping with the mathematics ; the shift indica tes a way of thinking about systems behaviour in which the featu res of the models are deeply linked to the systems they are descri bing. So, in the case of our electronics or control engineer, the poles effectively cease to be just convenient visualisations of mathe matical complex variable theory, and become system features which are just as real as the electronic components from which the

Models and black boxes

333

system has been built. Explanations of physical behaviour also shift in the same direction ; at a systems level accounts of how and why a system behaves as it does are often couched in terms not of phys ical variables such as voltage and current, but in terms of where the poles of the model lie. In this different ontology the system poles are not simply part of a convenient mathematical model, they are what cause the system to behave as it does, as if they had the tangible existence of the physical components and measurable signals present in a system. From an engineering viewpoint, then, this language is a powerful way not only of representing important aspects of systems behaviour, but also of explaining how that beha viour comes about (48). Similar points can be made about other applications of mathe matics by communications, control and electronics engineers - the language of spectra, the language of correlation (as indicated earl ier) or the language of constellations in signal space in modulation theory (49).

IX. Conclusion The main argument of this paper is that communications and control engineers developed a radically new approach to the modelling of linear systems, an approach that has been underresearched (indeed, largely ignored) by historians of both communic ations and control. From the end of the last century, communications engineers ini tiated this new approach, first through the use of phasors and then through increasingly sophisticated time- and frequency-domain modelling. The abstraction of system components led to a meta language in which the manipulation of circuit configurations became a natural consequence of the mathematics. This approach encouraged engineers to concentrate on the input-output behaviour (48) A more extensive discussion of this argument can be found in Bissell and Dillon, 2000. (49) Mathematically, a signal constellation is a set of points in a vector space, but is treated by engineers in just as concrete (and perhaps, to mathematicians, as cavalier) a fas hion as spectra, poles and zeros ! Communications engineers will talk about placing the points in signal space, about how the distance between them is relevant to performance in the presence of noise, and so on.

334

Chris Bissell

of devices and networks, thus leading to black box and sys tems ideas. Linear systems ideas that originated with communicat ions engineering were extended to other domains, such as control engineering, although the manner in which the transfer of ideas took place was not straightforward, and was a result as much of social as of technological influences. Communications and control engineers developed a new way of exploiting these models, in which abstract concepts such as poles, zeros, spectra, or signal constellations took on an almost palpable identity, and were manipulated in a highly concrete fashion. An extremely significant synergetic relationship developed bet ween the mathematical modelling of systems, and the design of in struments and computers to aid design and analysis. The key notions of orthogonality (at the heart of Fourier analysis and linear systems theory) and duality allowed engineers to work simultaneously with equivalent models of different types. One of the most striking fea tures is the way that the modelling approaches directly influenced the design of instruments such as the signal analysers and correla tors that exploited these models. And, increasingly, devices and sys tems could be implemented that were very close to the mathematical idealisations ; the analogue computer marked an important stage in the development of this approach. Today, particularly through the use of information and communication technology, highly abstract and complex mathematical models are routinely implemented by design engineers without the need to engage directly with the models themselves ; interestingly, such contemporary engineering practice thus calls into question the historically privileged position of conventional mathematics in engineering education.

Acknowledgements Earlier versions of parts of this paper were presented at the Conference on the history of telecommunications 2001, St John's, Newfoundland, 25-27 July, 2001 and at a seminar at the Centre Alexandre-Koyr, Paris, March 2002. The author grate fully acknowledges the perceptive comments of Stuart Bennett, David Chapman, Amy Dahan-Dalmedico, Chris Dillon, Mike Meade and Gaby Smol on various draft versions of the paper.

Models and black boxes

335

Bibliography Anon 1996

The magnetic force of Charles Proteus Steinmetz, IEEE Power Engineering Review, 16/9, September, 7-12.

Atherton W. A. 1984 From compass to computer (London : MacMillan Press). Belevitch Vitold 1962 Summary of the history of circuit theory, Proceedings of the Institute of radio engineers, 50/5, 848-855. Bennett Stuart A. 1993 History of control engineering, 1930-1955 (Stevenage : Peter Peregrinus). BlSSELL Chris C. 1986 Karl Kupfmuller : A German contributor to the early lopment of linear systems theory, International journal of cont rol, 44/4, 977-989. 1992 a Six decades in control : An interview with Winfried Oppelt, IEE Review, 38/1, 17-21. 1992 b Pioneers of control : An interview with Arnold Tustin, iee Review, 38/6, 223-226. 1994 Spreading the word : Aspects of the evolution of the language of measurement and control, Measurement and control, 27/6, 149-155. Bissell Chris C, Dillon Chris R. 2000 Telling tales : Models, stories and meanings, For the learning of mathematics, 20/3, 3-11. [Pre-publication version at http://telematics.open.ac.Uk/pub/reports/2.pdf.] Black Harold S. 1934 Stabilized feedback amplifiers, Bell system technical journal, 13/1, 1-18. Bray John 1995 The Communications miracle (London : Plenum Press). Brown Gordon S. and Hall Albert C. 1946 Dynamic behaviour and design of servomechanisms, tions of the American Society of mechanical engineers, 68/7, 503-524.

336 Bussey Gordon 1989

Chris Bissell

Wireless : The crucial decade, 1924-1934 (London-Stevenage : IEE).

Campbell George Ashley 1922 Physical theory of the electric wave filter, Bell system technical journal, 1/1, 1-32. CAUER Emil et al. 2000 Life and work of Wilhelm Cauer, in Proceedings of the 14th International Symposium on the mathematical theory of net works and systems, Perpignan, France, 19-23 June. [Consultable sur http://www.univ-pe.f/mtns2000/articles/SI12_l .pdf.] Cauer Wilhelm 1926 Die Verwirklichung von Wechselstromwiderstnden vorgeschriebener Frequenzabhangigkeit, Archiv fur Elektrotechnik, /4, 354-399.

Fagen M. D. (ed.) 1975 A history of engineering and science in the Bell System : The early years (1875-1925) (Murray Hill, New Jersey : Bell Tele phone Laboratories). Fischer Claude 1994 America calling : A social history of the telephone to 1940 keley-Los Angeles : Univ. of California Press). The waveband theory 125/3142, 92-93. of wireless transmission, Nature,

Fleming Ambrose 1930

Foster Ronald M. 1 924 A reactance theorem, Bell system technicaljournal, 3/2, 259-267. GUILLEMIN Ernst 1935 Communication networks, vol. II (New York : Wiley). HUGGINS W. H. 1977 The early days - 1952 to 1957, IEEE transactions on circuits and systems, CAS-24/12, 666-667. Hugill Peter J. 1999 Global communications since 1844 : Geopolitics and technology (Baltimore : Johns Hopkins University Press).

Models and black boxes Huurderman Anton A. 2003

337

The Worldwide history of telecommunications (New York : Wiley-IEEE Press).

KPFMLLER Karl 1924 Uber Einschwingvorgange in Wellenfiltern, Elektrische Nachrichtentechnik, 5/1, 141-152. 1928 a Ober Beziehungen zwischen Frequenzcharakteristiken und Ausgleichsvorgngen in linearen Systemen, Elektrische Nachrichtentechnik, 1/11, 18-32. 1928 b Ober die Dynamik der selbsttatigen Verstarkungsregler, Elekt rische Nachrichtentechnik, 5/11, 459-467. 1977 ntz-interview, Nachrichtentechnische Zeitschrift, 30/10, 774776. Lee Y. W., Cheatham T. P. Jr., and Wiesner J. B. 1959 Application of correlation analysis to the detection of periodic signals in noise, Proceedings of the Institute of radio engineers, 38/10, 1165-1171. Mindell David A. 2000 2002 Opening Black's box : Rethinking feedback's myth of origin, Technology and culture, 41, 405-434. [http ://web.mit. edu/6.933/www/black.pdf] Between human and machine : Feedback, control, and computing before cybernetics (Baltimore : Johns Hopkins University Press).

Nyquist Harry 1932 Regeneration theory, Bellsystem technicaljournal, 1 1/1, 126-147. O'Neill Eugene F. (ed.) 1985 A history of engineering and science in the Bell System : mission technology (1925-1975) (AT & T Bell Telephone Laboratories). On the role of mathematics and mathematical knowledge in the invention of Vannevar Bush's early analog computers, Annals of the history of computing, 18/4, 49-59. General-purpose electronic analogue computing 1945-1965, IEEE Annals of the history of computing, 15/2, 8-18. The Analogue Alternative (London : Routledge).

Puchta Susan 1996

Small James S. 1993 2001

338

Chris Bissell

Solymar Laszlo 1999 Getting the message : A history of communications (Oxford : Oxford Univ. Press). Steinmetz Charles P. 1893 Die Anwendung complexer Grssen in der Elektrotechnik, Elektrotechnische Zeitschrift, 42, 597-599 ; 44, 631-635 ; 45, 641-643 ; 46, 653-654. 1897 Theory and calculation of A. C. phenomena (New York : McGraw-Hill). Wass A. A. 1955 Introduction to electronic analogue computers (London : Pergamon Press). ZOBEL Otto J. 1923 Theory and design of uniform and composite electric wave fi lters, Bell system technical journal, 2/1, 5-39. 1924 Transmission characteristic of electric wave filters, Bell system technical journal, 3/4, 567-620. ZVEROV A. I. 1966 The golden anniversary of electric wave filters, IEEE spectrum, 3/3, 129-131.