You are on page 1of 72
int on the mac all its transitions: hed as to whether or ® fauit-detection expert iment) ament the 3 ; an initial (or start f mth experiment the machine is said to be in « final stale. to distinguish between two types of experiments: 1, Simple experiments, which are performed on & single copy of the machi : 9, Multiple experiments, which are performed on two or more iden- tical copies af the machine In practice, most machines are available in just & single copy, and there- fore simple experiments are preferable to multiple ones ‘According to performance, experiments are classified as: L. Adaptive experiments, in which t put at any instant of time depends on the previous outp 2 Preset experiments, in which thy i ce is predeter- mined independently of the outcome c experiment. A measure of the efficiene: which is the total number of input the execution of the experiment. Tn Chap. 10 we studied the properties o Ainguish between two nonequivalent stat if 8 and S) are distinguishable, then ¢ euiakatier Gg experiment of lengthn — 1. Wenow consider the more general a that of dentfying the initial or nal state of a given aad of distinguishing a given n-state machine from all other n-state ee which bave the same input and output alphabets. ee y and cost of an experiment 16 its length, 2 symbols applied to the machine dur f experiments used to dit We showed thal wtroductory example tes A, B an oe! ich may initially be in any one in Table 13: of Afi to input sequences he 13-2. ‘Knowing the output. seq O1 ar ; the input soquence 01 is always suficieat ton ith | state - op indicates that the final stat er ase that the final state ts uw t the response hand, the knowledge of the respon’ " ‘On the other ee le determine sani! : input seq en ee OL is mot feces GD could mean that senervind ee ee: s initially in either ‘was A or that itwas 8. In fact, if A was teen Ree is impossible to determine the initial state by oy ; aa atarte Miwa, since the ent of a eres ie : i uent “and the output produced in both oases is a io aa Pane initial O input will yield uny new information regarding Table 1 Machine Ar, Using the same line of argument, it is evident that Mauence that Jf; produces in response to t cient to determine uniquely Ms final state, as well a8 its initial stute, shown in Table 13-2, each of the output sequences that might result the application of 111 to My is associated with Just one initial state hd one final state. Before presenting the techniques to be used in pats, we shall introduce some of the terminology » Which will prove to be an effective tool in Eperiinents the ow tput. he input sequence 111 is always f the design of experi- and define the Successor the design of minimal Table 13-2 Re: ses of Mi to input sequences 88 and 111 Initial Response ia Final Anitiat Responae to state on ate Finat state 1 “ ee tea i oa a 4 110 B 00 8 8 ML ig. "1 B e ou p a1 4 2 101 ig the state of the machine is given b uncertainty is the minimal subset of 5 s known to contain the initial state. For example, if ‘tially be in any one of its four states, then the initial Fei aiet in to perform experiments that reduce'the inital and whenever possible reveal the initial or final state. For ex, suppose we apply an input t to machine A, and in response it prod an output 0. We may eonelude that Jf, has initially been in state ¢ since only from that state is a response of 0 toa L input possible ‘Thy final state in this case is B. However, suppose the response of My te an input | is 1; then all we can say regarding the final state of the maching is that it may be any of the states J, A, or C, depending on whether the initial state was A, B, or D, respecti The set of states (ACD) thus represents the uncertainty regarding the final state of My after the appl. cation of a Linput. Ingeneral, the uncertainty regarding the state of Mf after the application of X is a spe ibset of the X-suecessors of the states contained in the initial u tainty. The elements of the uncertainty are not necessarily distine Let Uy be the initial uncertainty, and let the input J, result in an uncertainty U,; then U, is said to be the J,-swece of Us. Suppose, for example, that the initial uncertainty recarding the state of My is (ACD). Jf an input 1 is now applied to M,, the successor uncert: ainty will be (BY or (CD), depending on whether the output is 0 or 1, respectively. We thus say that uncertainties (5)(CD) are the L-successors of (ACD). Subsequently, we shall refer to ction of uncertainties a3 an uncer tainty vector. The individual ur inties contained in the vector are called the components of the vector. An uncertainty vector whose com te pee ae each is said to be a al uncertainty vector. idaatical Sead ee ae nts contain either single states oF Stihiis for ecole the bra th honiege neous uncertainty vector. ee a ts ‘ a AA)(B}(C) and (A)(B)(A)(€) are homo @ graphically the J,-successor unc ie the experimenter in the sclection of theme 1S composed of branches arranged in suc ‘og. The highest node (in level 0) is asaneio aenty Ue. and each of the p nodes in level Lis an. of Us. The jth level of the tree comes or a ctingatanode. Asequence of j branches, ee vend terminating at a node in the yth level, is a free; J 5 called the length of the path. Each path equence which, when applicd to chine, results yeetor gssuciated with the terminal node in the jth eves with j + I levels contains p" paths, describing, the of length 3. oe tree for machine M, nnd an initial unc Tt contains four el ity (ABCD) 0 through 3 is show in Big ET “a laacol 0 1 tailgce o 1 2 2 (alles lags) (aliaciicd ane 0 3 > (aANCHIC) (aiaiane) — sallaneric) Lalanentor Fig. 18-1 Successor tree far machine My TS labeled with the input symbol which it represents, and associated with the corresponding uncertainty vector, The Bsiociated with the initial uncertainty, while the nodes in, Oeiated with its 1- and (I-suecessors, and sa on. For exam ,” M, when the initial uncertainty is Ec fertainty vector (ACD)(B), while a 0 input results in (ABCC). ‘The tascconor of the vector (ACP)B) g the I-successors of (ACD) and (B) separately. es8or of (B) is (A), since the application of a4 State B takes it to stale A. ‘The I-successot ‘of (B) is (C), while that of € 4 ‘uncertainty 13 said to be smaller thar. ty fewer elements; eg., (BC) is ‘smaller than ( inty the tree is constructed, it is evident that an uncert® OT "1 same with a node in the jth level is either smaller than or com vel A homo- Srobitier eC elemanta ax ita predaseseor in the (7 — Ut eee cous geneous uncertainty vector will always have ss its sucoessors VV the suc- ‘uncertainty vectors, Wor example, in the tree of machine “ol , Bengt Gl Ue unceriainty (HCC) wre (A.A)(C) end (A)(BB) 7 7: : actiea! value, may be continued as far as is neceadary, but for it to be of prablin ? truncated version must be defined by stipulating a number o Sir tion rules. nod (A) 13-2 HOMING EXPERIMENTS ing nbjective of this section is to develop techniques Tor the construc- tivn of experiments to identify the final state of a given n-state machine. Tt is shown that sueh experiments ean be constructed for every reduced machine, aid bounds on their lengths are derived ree if om. the Definition 13-1 An input sequence Vp is suid to be a homey the final stute of ihe machine ean be dete muchine’s response Lo Vp, regardless of the initi ned uniquely I state. The homing tree A homing sequence for a given machine M may be obtained ¢ tuncated version of its successor tree. Our task is to construct the ter and to determine the shoriest path leading from the initial arene to a trivial or a homogeneous uncertainty. The prosenee at ne! uncertainty at the kth level of the tree guarantees that i of such an ‘input sequence consisting of k symbols whose aaueie there exists an cient to specify uniquely M's final state. ion to AY is suff. A homing tree is a successor tree in whi "terminal when any of the following occur: St) evel node becomes ve itis identical wath node (ABCD) in level | is ter ABCD) in level 0. ‘The nodes in level 3 are terminal nod: D)(D.B) is «homogeneous uncertainty vector. The shortest f quence is O10, since it is the shortest sequence described by a path . devel a (agcoh posh ee (asco) 2 (ashoa) me e ib 3 [aloneo) laailac) Fig. 13-2 Homing tree for machine Mf, from the zeroth level to a homo; IRENEOU and final states corresponding to Seren ee this sequence are given in pee the existence of the homing experiment and 5 ror near anets howe length is af most ae ion of the sequence Aa n Bie ar pack the A,-Succestor uncertainty vet, i tains at least two components. Next, select any two Sue component and apply the appropriate sequence hs, which Ainguishes between them, The AiArsuccessor uncertainty . contains at least three components. In a similar manner We obtain the Aide - © * An-s-Successor vector which consists of n each of which contains only one state. Therefore the sequen ahs “h, is a homing sequence whose length is at mos (n 4 Re of Bf; to ‘the homin nee OLD Initial —Responacta, Pinal atate O10 state 4 on B 001 e 101 D 101 This value is an upper bound on the length of the homing sequence, but it it upper bound. Tt can be shown that the length of the homing se > need not oxce ( — 1) and that this is indeed & tight bound Prob. 13-4). Tho proof of such a tighter bound ie beyond the seape of this book proof that an adaptive homing experiment, whose length is at me 6 1), exists for every reduced a-state machine is given as an Synchronizing experiments A-synchronizing sequence of a machine M is a sequence which takes M1 4 specified final state, regardless of the output or the initial state. Some machines possess such Sequences; others do not, For a given machine we can Meek: A jitleval node of the following occur; with an uncertainty thatis also associated in a preceding level, he gth level is associated with an. Uncertainty a single element. The synehro- ae lage (ce) 4 Me D Syechronizing tree for eetuence is described by a path in the tree leading a Y fon singleton uncertainty, i.c., an uncertainty s oe For machine Af, the path ‘01010 deseribes use which, when applied to Afs, synchronizes the Heardless of the output or the initial state. Note nty of machine M, is (BCD), the sequence 010 Dy since the 010-successors af B, C, and D are ; be. and apiply to them s sequence £1 ¥ ‘some state S,. This task can always be aceouy ' is known to possess a synchronizing peaisceiae PER the sequence ¢, is at most (n — 1)n/2, sinee the longt the synchronization of (S,S)) is through all possible pairs es iS), (18), 1, 8.8). Pramas : cessor of (8,5.). Next select a state 5, ‘ tainty, and determine the sequence fs, which takes (SiS,) into SE state S,, The length of f, is also at most (n — 1)n/2. In the CAE way it is possible to find the sequences fs, £4,» - + Exar SUI when concatenated, yield the synchronizing sequence ffs ~ °° fot whose length is at most (x — 1)71/2. Sy is the §-suc- from the resultant uncer The above bound is not the least upper bound; in fact, the least upper bound is unknown. For a tighter bound see Appendix 13-1. 13-3 DISTINGUISHING EXPERIMENTS: Distinguishing experiments are concerned with the identification of the initial state of a machine whose state table is known, but no other informa- tion regarding its condition is known. Preset distinguishing sequences Definition 13-2) Let AY be an n-state machin An input sequence X, ig said to be a distinguishing sequence if the output sequence produced ‘by 5 Af in response to X, is different for each initial state, e Knowing the output sequence that Mf Produces j is sufficient to identify uniquely M's initial state mice lo X, _ the initial state and the input sequence is always sufficient nowledge af uniquely the final state as well, Consequently, enon, 9° 9° sequence 13 also a homing sequence, a ¥ dit The converse, ho many homing sequences do not provi ha eet is not 4, the initial ea Provide all the rm; Tue, +» the sequence 010 for machine * ‘ation regard. e 4 r 7 Falta naacetatod with « trivial ubeertAl™y inguishing sequence of A if and unly (which is assumed to consist of ie sociuted with at inguishing sequences I= ates in a nod length of di AL, is obtained from the coree- node associated with the home- or (AI(BCE) is terminated, sir no further Le., there is no way of knowing, whether the initial state was A hing sequences of length 3: ULL, af AF, to the sequenee 1 LL is r distinet respo MEOUS UNcestainty y BPEFIMENE Can spliL thy mee the machine has TE. Machine 101, and 100. The respons: T: 13-26, This epending on the init While every 1 distinguish eM, imust be t mated tn Le itial uni compar would have revea va O will distinguish betwe A and #, while no experiment which starts the initial uncertainty tainty, and the ve trivi wchine At: ( [t, since no experiment which starts wi tes C and D or t fith » 1 will reduc nt. An ing tw inguishing prefix shortest di: many eases the initial state of a machine co rmined from only fix of the distinguishing sequence X,. The length of the required x is a function of the initial state. Consider, for example, machine whose response to the distinguishing sequence 111 is given in Table Tt is evident that if the response of the machine to the first input 1 is Q, the initial state must have been (’, and the distinguishing nt may be terminated at this stage. On the other hand, if the js 1, the initial state could have been either A, B, or D. The at must continue, and Af, is supplied with the second 1 input, » ig now Q, the initial state must have been D, and the be det f tree (See Tt ; r ‘of fault detection and machine identifie, poet ine a eatvely short experiments. “Adaptive distinguishing experiments ‘ So far we have considered preset experiments in which the choice of eaph input symbol is predetermined and is not influenced by the response of the machine to the preceding input symbols, We now consider adaptive experiments in which the choice of each input is determined by the ‘machine's respanse to the previous inputs. As we shall see, the advan_ tage of some adaptive experiments is that they are relatively short; in s addition, there are machines for which only ac constructed. On the other hand, the main disad tive experiments is the relative difficulty in d to inspect the output, after the applics ‘ tends to slow down the experiment. As an example, consider machine M, ( | for whieh the shortest preset distinguishing sequence is of length $. Supposé that the Table 13-5 Machine M, nt, uneerixinty is (ABCD) and an input symbol 0 is applied, PAE Been a acertainty is (CD) or (BC), depending on whether the response Gor, respectively. This process can be described by me adaptive tree, as shown in Fig. 13-4. In this tree, each horiz9 the application of a specific input symbol, a3 ® label. The downward lines represent the hine can produce with the given ut i fa) A partic! adoptive tree (aco) ee a} mA Db) A vortil vee describing the ting an adapt fig. 13-4 Constr tinguish periment. fer ine Ms. Input symbol that must be applicd to the machine after the first 0 depends on whether the machine produced f, that is, whether the unccrtainty is (CD) or (BC), effect, each of thi un be viewed as fora new experiment. If the uncertainty 18 (CD), linguish state C from state D, a | input is appropriate; ADinput will distinguish between states # and C. In E adaptive experiments, each nontrivial component tainty vector represents the observer's knowledge the machine, and consequently can be viewed a8 uncertainties specifies the input symbols in the course of am ta be 0 distin « symbols. E : wilt - experi : alate ee eee oe ae inet tae experince PNT Yo an Be inctane Fe ta: ent that noisuct Bom Fehe adapeve Hale on a sin me a Thus, in designing Pe ries wrlainty containing repeated entries. repeated ents neertuinties. is terminated and so are the paths that evcairrngts this ond Our objective is to obtain minimal adaptive experts i symbol that will Wwe shall choose at each point in tho experiment an imPulSyO ug. minimize the length of the path leading from the point tos experi i ll have an adaptive distinguishing Pa ive tree contains a sct of paths which ¢ ee inty, such that cvery path leads to a sing ‘The adaptive distinguishing experiment for machine M, is given by the partial tree in Fig. 13-46, where ) of the paths emanating from the initial uncertainty (ABCD) leads to a singleton uncertainty. ‘The experiment requires two input. symbols, while the shortest preset experi- Ment for this machine requires three input. symbols ¢ also that since each of the paths in the experiment is minim: i __ preceding definition, t! e adaptive experiment is minima experiment is summarized as follows ceording to the The complete 1. Apply an input 0 and inspect the corres Bic s ieeiep ipe rae ie ti output. If the 2 Apply an input t. The initial state uniquely by observing the response was B; if it is 1, the initial state was 4 | Apply an input 0. 1 the res state was C. pea can now be de If it is 0, the ini termined tial state ore © to this j, (itis 1, the initial state wag pont % the initigy appl stat And tedu unig itsel concerned with tho problems of identify’ ee a known machine. We shall now addre ing an unknown machine, 1 f experimentally, Ne ee eee eee eee c acncral n i ss most general tining the state table of an unknown machine. In. its mo ino, the when no information is available on the unknown machine, ue jeannot be solved for several reasons. Virst, the expe te : ist have complete information regarding the input alphabet of th fine, since otherwise he or she ean never be sure that the moxt inpul not reveal new information regarding the machine, Simi- + the machine cannot be identified unless there is an upper bound on umber of its states, since for any given machine and xy is possible to construct another machine that responds to experiments of length & exactly like the given machine, but ond differently ta experiments of long fugiven machine AM, is in a tally from machine M; whoso initial state ough machines Af, and M, may, in fact, b n clearly will not ocour if both Af To muke the problem of machi wal restrictions on the machines, mewn, as is an upper bound on the number of states of the machine ver, the machine is assumed to be reduced and strongly ron | An unknown machine with at most n states Can now be ider @following manner. Construct tho direct-sum table (see Pr Pall the tables that have « or fewer states and find Clearly, such » homing sequence can always be found, and its ation to the machine in question will reveal which s¢ from the direct-sum table contains the final st ate if the direct-sum table contains only those tables th: d and strongly connected machines, the ely identify the final state of the machine and, in turn, the machine ~ This demonstrates that under specified conditions the machine ation problem san be solved. However, as a Procedure for ng experiments, the Timent of pe distinguishable homing 5 f equivalent of the machine, at correspond to 2 homing sequence. will The first step in the analysis of these sequences is the identification ‘of the distinct states of the tested machine. Let us name the two states ‘A and B and suppose that at the start of the experiment the machine was im state A. The application of a 1 input results in a 0 output and a transition which is yet to be determined. However, since the second input symbol is also a 1 but the response is 1, the machine must have been at fin nstate other than 4. Hence the experimenter may conclude that at ¢: the machine was in state &. Since state A is the only state which responds to a | input by producing a 0 output, it is evident that at f the machine was in state A At i, it was again in state #, since it has already been verified (at 4) that a Limput causes a transition from state A to state 8, Ina similar manner it is easy to show that at f, the machine was eB, which in turn implies (see ts) that at f it was in state A, tf; it must have been in state A, since this is the only state in which the machine produees & 0 culput as a response to a L input. As a result of the above analysia, the experimenter is able to demonstrate that the machine indeed has two States, named A and &, and that its transitions and outputs are given by the state table of Table 13-6. Thus, the above experiment is an identi- fication expcriment for machine M, Table 13-6 Machine Af, of designing fault-detcction experiments (or checking €1 u a restricted problem of machine identifieation. ? our attention to strongly conne ree machines. We also assume that the _ ome component failure. ‘This assumption to noise oF incorrect inputs. At first we at least one nguishing sequence, all relax this restriction and discuss machines ence, Note that the eriments are faults, but will not locate : ‘presence of one or more n experiments ‘we shall use, each fault-detection experiment consists an adaptive experiment. whose aim is to transfer arat part is : the mu chine into & prespeci fied state, which is the initial state for the second part of the expe pont. 2. The second part is a preact experiment in which the machine is 1 urt is subdivided through all possible transitions. ‘This into two parts, In the first part the ma chine is play the response of each of its states to the sequence, svhile in the second part tt anal tr verified, ed to dis= nguishing itions are consider machine Af,, whose state table is given in responses to the sequences 00 and O1 are sum- #138. Suppose that the preset part of the experiment Sinte A is the initin! state, to which it is therefore Table 13-7 Machine Af, Bee Cto least four | states, ther the disting if at any t responses t Retessary to transfer the machine. To this end we apply the homing machine at Sequence 00 and observe the response. that the na ‘ of mares wi 1 Tf the response is 00, the machine is in state €. Apply the TIE the Wansfer sequence} T(C,4) = 00 to transfer tha machine to stain | point, we an A; without ap: 2, Wit isO1, apply T(D,4) = 0, been prodiy $ If itis 10, apply T(R,A) = 10, has operate 4 Iie ine is im state A, actual final In designing the preset part of the fuult-detection experiment, the machine ac first task is to ascertain that the sta te is inderd A and that the tion is ineg machine being tested actually contains four d es. This can be To co accomplished by displaying the resp tinguishing sequence. Muchine Af, has tw 00 and 01, whose applications to the machine res im Table 13-8. The design of exporin Sequence 00 is somewhat shorter, but w exercise, To display the response of t tinguishing sequence X, = 01 state trans input symt applying we shall © same dis- ishing sequences, ts shown - distinguishing ill be left ta the rey pply the di uted correctly’ and it is now in state D. To displ euishing sequence O1 is applied agai ‘he application of a thi Machine in state B and displays thea twice more leaves the machine in state If the machine has ope ‘a transition from existence of at has no more than four nee, and its response to From this point on, in bove ment one of the 2! rida, the state of the (it must be emphasized rtanee; a different set ¢ distinguishing sequence has been st any time during the course of the ex sses to the distinguishing sequence is PI at that time is uniquely identifiable. the names given to the states are of no impa names would result in an isomorphic machine.) Tf the machine has not produced the expected output up to this it, we may conclude that a fault exists and terminate the experiment out any further tests. If, however, the above expected output has Produced, no eonelusion can be reached as to whether the machine ‘has operated correctly and is indeed in state B or a fault exists and the actual final state ix different from B. We therefore assume that the | machine actually started in state A and terminated in B. - If this assump. correct, it will be revealed in the next part of the experiment | To complete the experiment, it is now necessary to verify every iets transition. ‘The general procedure to be followed is to apply the GpmPut symbol which causes the desired transition and to identify it by lying the distinguishing sequence. ss Shall start by applying an r ped the machine is in state B, " input 0, followed. by di Quence O1, This input sequence takes the Rae tae to mae nd thus « 101 input is applied to check the transition from B to Di ® Jer Tinput and to verify that the machine actually has moved to at: sal each of these three-bit sequences, the first bit causes the aa sae the distinguishing sequence ascertains that the transition is sn At this point we have obtained additional ining sa out another transition. It has earlier been shown that th BOL to the machine while in state B causes it to go th stat, ince an input of 0 takes the machine from B to formation Prlication or of B is also C, the l-successor of C must be C _ At this point the machine is in state C, If, inv (001, the machine produces an output that the O-successor of C is D, and has already been established that in response to the input Sequence 111, we ‘ths final state ie paca ; have checked every possible transition, eee aie and fom A to B aad to C. Since the machine uly in state C, we must apply a transfer sequence to get to either aD ic "A, Such sequences can always be found for strongly con. ‘nected machine, and require at mast » — I symbols. Furthermore, the transfer sequences should be applied in such a way that they will take the machine through “‘cheeked’ transitions only. ‘Thus the only possible iransfer sequence in this case is T(C,0) = 0, because, us has already been demonstrated, the machine goes from C to D under an input © The ‘application of a 0 followed by O1 ascortsins the transition from D to A ‘and returns the machine back to state D. TI provides enough Let 51,52, machine M is information to verify the transition from A to a linput. This XT(Q5: verification is achieved by inspection of the preceeding nce and will serve to ta observing that Cis the Ol-suecessor of D, and A is the 0-suceessor af D. @ diffe ‘Thus C is the L-successor af A ‘The last transition that needs B. Since the machine is in stat applied, followed by 001. ‘The complet 1 is from state A to state squence T(D,A) = Oia shown below: to Xs, when similarly lod th Xs is applied fo few O1Giotototaa The preset part of the fault. ‘the above input sequence, whose length is 27 symbols. If the machine ‘at hand responds as shown above, it must be isomorphic to M,, since it ‘has been shown to contain four states whose responses a corresponding responses of Afs, and all state transitions, which " ‘Verified in terms of behavior exhibited at the beginning of the experiment ‘are also isomorphi¢ to those of M, Clearly, if the machine has not pro ~ duced the above expected output, it cannot be operating correctly. ‘The location of the fault, however, cannot be determined merely by the above first. E Thus, the nexs jn cach case the transition is verified by dis- distinguishing sequence. In practice, it is not the responses at the beginning of the experiment, tion that is verified at a Inter point in the experi- idotermine & state transition at some earlier point, jy, the procedurc for constructing fault-detection ines having distinguishing sequences is as follows: iS. be the states of machine M, and suppose that X, sequence for this machine. Let a be the state to ially in S, as a result of the input sequence Xe fan input sequence (not necessarily unique) that ie 1 eh ete to state Now suppose that — tially in its starting state $; Then, the sequence VeF(Gs.50) distinguishing *9 which Mf goes when init KMS IXeTS will sorve to take the machine th all the different responses to the starting in Si, Xo leaves the machine in @ the machine to &, where X, is applied again, leaving the mac The corresponding output seque cle: to Xa, whem initially in ers similarly led through all its ne states, a ‘Xv is applied followed by the transfer ugh each of its states and to display tinguishin For example, ) transfers na point t (Q.,8: ent th: sequence XyT(Q..8,). Uf it operates correctly ‘This is verified by applying to it again the distinguishing sequence Xe. Clearly, fit tesponse to the last X, is identical with its response to the first Xs, the machine w leed be at the end of this part in state Qu. ‘Thus, thenert part of the experiment can start at this point by identifying ‘wanatons out of state Q,. Part of the experiment we establish the various state HAO shetk, for example, the O-transition out of state i, ingome state Qj, the appropriate sequence is Xa guarantees that the machine indeed went fas done in the.previous part of the experiment. to state §,, and then OX, is applied to cause ut of §,. In a similar manner the machine can ble to apply the two parts of the experiment simultancously ad of sequentially. The method outlined above can be applied to any redueed and strongly connected machine that has at least one distinguishing sequence. ‘The design of fault-detection experiments far machines which do not have any distinguishing sequence is studied in Src. 13-8. As will be shown, the design of such experiments is extremely complicated, and the resulting experiments are very long. This situation leads to the conclusion that, when designing a sequential machine, care must be taken to design it in such a way that whenover possible it will possess som distinguishing Sequences and, thus, will br simpler to maintain and test. The develop- ment of such design procedures is the subject of the next sretion. ‘Th ‘The row ‘1 DESIGN OF DIAGNOSABLE MACHINES are thein A diagnosable seq uci machine is one which pass one or more dis- column J Unguishing sequences and thus permits us to identify uniquely the states be of the machine by inspecting its uch 8 In this the eit Section we shall pi & method to modify the nitial ‘sorrespo Machines in such a way that they will po shing some pil Sequences so that relatively short fault-det an be some ed constructed for them column , is a dash complet ‘The testing graph Wi Machine My, (Table 13-3) does not possess any distinguishing sequence, sor (8,8 ‘We shall now show how it may be augmented, by adding to it an addi. by (AB: tional output terminal, so that the sugmented machine will possess several is, all eee ing sequences. ‘f state table of Af; m: rewritten ax shown in the uy of Table 13-9. ‘The column headings consist of all input-output ae _ tons, where the pair 1/0, corresponds to combination of i i sutput 0), ‘The row headings in the uppe Si, is the fy-suec put O; and isa dash ¢ : is B, and the : 8 Bis entered in row A, elon nit entries of Mf, are F hi Fow hesdings are all unordered pairs of states, while the table entri their corresponding successors If the entries in rows 8, and plumn /4/,, of the upper half are ly, then the ent row 8,8,, column f,/0), of the lower half is S,S,. For example, sinc i pectively, the on. If for Dair of states 8, and §, either one € column 44/0, are dashes, the nding entey in lama I,/O;,isudash. For example, the entry in row AC, © ja dash, since the entry in row C, column 0/0, ix a dash pleted is referred to as a testing table We sball refer to a pair (S.5;) ns an uncertainty pair and to its or (5,S,) a8 the implied pair, Thus, for example, the pair (#D) (AB). An uncertainty pair that docs all the entries in the corresponding row from the table. Whenever an entry ited state (ee., DD in row CD), d DD means that states Cand D and are indistinguishable by an implied imply any other pair, that are dashes, can be omitted the testing table consists of a that entry is circled. are merged under input. 0 into state “speriment which starts with a 9 Tet us define a directed graph @, which will feller: will be called a testing }. Corresponding to exch i - Es there is m vertex in (7. Tow in the lower half of the testing table 2. Uf there exists an en| 5 s in row 8,8, column, arc leading from machine Af; is derived directly from the ‘and is shown in Fig. 13-5. Fig. 13-5 Testing geaph for machine Ms Definitely diagnosable machines ‘A machine M js defined as a definilely diagnosable machine of order a if # iis the least integer, so that every sequence of length » is 0 distinguishiag Sequence for M. In other words, a machine is definitely diagaosable if every node in level w of the distinguishing tree is associated with # trivial Uncertainty vector. ‘The distinguishing tree can thus serve as a tool fot L er izing definitely diagnosable machines. In this section, however ll man a different test by means of the testing graph. “A machine M is definitely diagnosable if and p ie eer fore de coment i fa pot contain repeated entries Now — Joop-free. ‘Then, by repeatedly applying with the labels of the arcs in the loop, we feng input sequence that cannot resolve the the initial state. Consequently, the machine 7 Miagnoiable. ‘To prove sufficiency. assume that @ eit A is not definitely diagnosable, then there exists S fyag path in G corresponding to some input sequence H pair of states S.Si, s0 that cannot be distinguished eee But since the number of in G cannot exceed 2 rt nding to the number of distinct pat tates), (rarity long path in @ are possible only if it contains ‘Phuis the theorem is proved. = ‘The abowg testing procedure is clearly equivalent to testing by ‘of the distinguishing bree. In fact, the graph b ed with an ing loop-free ertainty weet me whose nonhomogeneous comp! with some node in a preceding level. Similarly ing table is free of repeated 1 with ar tainty ‘entries, it means that no node in the tree vector containing & homogencous nantrivy al ce node in the wth level of the tree is associated wi veetor be of repeated entries, Corollary Let the tesling table of machine M cand let G be a loop-free testing graph for M. If the le ngth of the fongest path in @ ta i, then p= i + x Since G is loop-free, Af is definitely diagnosable, Assume #>1+1; then there exists at t one uncertainty pair the application of an input sequence $,5,). Consequently, there must and S,S, in G, whose length yption, and thus » cannot exceed aller than J+ Lis trivial @ fh the general result that if a machine i definitely 4% then » > (n — L)n/2._ In Prob, 13-28 we show ot the least upper bound for #- shall in fact show that th nt to make M; definitely diaga ae end is to assign different outputs to each eo ‘a repeated entry in the testing table. In the eee nplished by assigning an output 10 to the ane ian essignment ‘and an output LI to the transition from D to D. at will be free of of output values ensures that the testing table of M: repeated entries. nad AB The testing graph of AM; contains three loops: a self-loop ANN and two other loops, each containing three vertices. Clearly, Sanh muat, be opened if Mf is to be definitely dingnosable. In general, 8 os is opened by the removal of any of its arcs. To remove an 4 ia necessary to assign different output symbols to the next-state entries represented by the vertex to which that are leads, In other words, an are leading from vertex 5,8, to vertex S,S, is eliminated by 2 different output symbols to the transitions from 5, and S; to For example, the self-loop around 4A in Fig, 13-5 is opened by assigning the output symbols OL and 00, respectively, to the next-state entries B and A in column z=0. The loop AB — BD.— AC — AB can be opened by the removal of the aro from BD to BC. This is achieved by ter assigning the ouiput symbols 00 and 01 to the next-state entries # and ( Th in rows B and D, column z = 1. Ina similar manner we open the loop of AC — AD — CD — AC by assigning 2 00 output to the next-state entry exe Din row A, column 2 = 1, thus removing the arc from AD to CD. Th resulting state table is shown in Table 13-10. ie : te le Table 1-10) Machine My Pres » of the entey in raw C, = the ae 2, und consequently the order of the modified 3. This renult can, however, be improved by specifying ry in row C, column z = 1,01. This specification actu- h Rat ures from AC to AD and from BC to AB. Asa result, the longest path in the graph is now 1, and M; is definitely ble. i 19-6.) THE 2 The distinguishing tree of machine My is shown (ascot (alisiionoy facitapl 0 i Laliaona) Laaveiot 4-124 Distinguishing tree for machine Af lear that, for a ficient to eon fp nal required aut definitely dia ing the nals is unsolve , although th imply the addition of n Rination of too many arcs does t Using pre is unavoidable Foutput termine Since the procedure followed in the above example can be applied rrive at the following general result: ny machine, we ro every reduced machine M there corresponds a definitely diag- Piable nachine A, which is obtained from 3M by the addition-nf | bne or more output terminals m of the definitely diagnosable machine M" whieh corro- kan AC is shown in Fig, 13-7. Is to machin [6] Mochine 4¢7 2, isu diognoting purposes. d only tor Fig, U7 Design of a definitely diagnosable machine, The question now arises as to the purpose of desiening definitely diagnosable machines tly, fault-detection experiments ean le designed with just one distinguish sequence. Moreover, even wheg & machine possesses two or more distinguishing sequences, there is mo known procedure for utilising th I ly in an experiment ‘The main motivation fo gnosable machines and studying their properties is the expectation that such machines will prove easier to maintain, and that it will be ble to design for them fault location experiments. Such exp will be simpler to design for definitely diagnosable machines, since it is possible to crass-check the machine with every sequence of length », and not with just a single Sequence, #137 SECOND ALGORITHM FOR THE DESIGN OF FAULT-DETECTION EXPERIMENTS ‘The main feature of the algorithm presented in this section is that eer the use of distinguishing sequences. with repeated symbols, & Sr H11, and thus allows overlapping portions of the distinguishing qeuvences, This in turn will result in substan ally shorter fault-deter enanments, For example, if 111 is n distinguishing sequence !@t on the addition of just a single 1 following the initial ¥ the initial (Clearly, if Af, is not AF, is initially sequence of length to transfer it to- In order to simplify the notation, we be by decimal numbers, ie., 00 wil and Il by 3. The fault-detecti distinguishing sequence wise of the initial state. of A is D; to display its a single L input is needed follow- we at this point consists nitial state, while ‘ent starts with the application of th X= T1 to the machine, 30 85 to di the ‘According to the state table, the I-sue fesponse to the distinguishing sequence ing the initial |. The input to the m: of three consecutive I's; the first two 1's the last two I's identify its L-suceessor If the machine is operating corcectly, | order to display the response of state € to th it is necessary to apply anather | inp Tl in A, and this un be ve point the experiment is as follows Boe bb Input: ee 1 State: ADCADE will be at f; in state C. In sequent will now be back n fiith 1. At this vs to the different ‘ und C (att). Since id {; are the same, must have been at states. Wedenote these states A (at 0), 0 ( e outputs produced by Af; in response to X, at ty a Verify in the course of the experiment that the oe BE ic distinct states, we may assume that at G the am ‘A. Moreover, since we know that the input a the machine from A to C, at ty it will’be in state C. In the responses of states A, Dand C to the distin- hove input sequence verifies the transitions from. ‘This sequence will leave the machine in stale the application of 1's is not aseful, since Dit 1 has already been checked. TI wed by 1. The 0 will Lake the machine to state & d by its response to 11. Then, to verify the transitiog: ler an input of 1, a | is applied, which leaves the machine j ‘An additional 0 followed by Xy results in the following: npul: Prt toeriog1i101t State: ADCADCDCAHBBBADE ‘Output: OLriorrtrriooooar 1 At this point the machine i: state C_ But since the two trang. 1 tions from state € have already been checked, a transfer sequence must, be applied, #0 that it takes the machine through only “checks transi- 3 tions. Since the only transition that has not yet been verified is from D to D under 0 input, @ transfer sequence T(C.D) = 0 is applied, q followed by O11, The complete experiment thus requires 19 symbols, { Peaeo ti orit.o.t toons Orzsziirivo00 01 2BBit 4 The general procedure The procedure employed in the preceding example is valid for any 1 f that has a distinguishing sequence with repeated it summarized as follows f 1, Apply a homing sequence and if ry a transfer sequence, #0 48 to maneuver the machine into the specified initial stat , denoted 5,. ea 2. iShooes Xe to be the Shorter sequence between the sequences ofl * all O's or all 1's. (For the purpose of clarity of the proc wit dure, assume that Xy has been cho: as the ull-1's sequence.) # Apply Xj followed by a 1 to check the first transition from S| The under an input of 1. (If the all-0's sequence has been chor, 3 apply an input 0 instead of 1, etc.) oan whicl 4. If the I-successor of S; is different f, ;, toon st J from S, yo Similarly, if the 11-successor of S, is different fr, ey acta sequi the I-successor of Ss, apply another 1. In the eater ot case, » continue to add 1 inputs as long as new Same manner 4 state k transitions are being When an sdditional 1 input does _ *pply an input 0 followed by: X,. As long as new transitions not yield any new transition i tions have t three steps until all transi ve ement for machine ‘The tion experemen! eee he snc sen Xo ON 1g isting! a is illustrated a5 follows x jerd ooo BBA ABBAS Zz: oo BOL ie F (b) Stop 5 Steps 3,4 1oo0100 a a ADDDCDOD i ein z: oe 3123 ie (c) Step 6 (d) Steps 7, 8 mM: OOO LTOLOOL0011LOD ABABBADDDCDDCAB z: roeorwooos sz 123111 (e) Complete checking experiment. wm #15-6 FAULT-DETECTION EXPERIMENTS FOR MACHINES WHICH HAVE NO DISTINGUISHING SEQUENCES procedures for designing fault-detection ich have preset distinguishing sequences much difficulty to machines which have only adaptiv: nguishing ences (see Prob. 13-28). The main difference is that in the latter different distinguishing sequences may be used to identify the various ites, while in the former ease the same sequence X. was used for stata tification. Moreover, adaptive distinguishing Sequences may be d even for machines that do have preset sequences, experiments for machines an be generalized without ¢ nes The use of aptive sequences will in most. such cases yield shorter experiments, } adap 1 tant tool in nts for machines Note however that, al-

You might also like