(4) complex coding for M-ary communication, and (5) adaptive equalization for MODEM's.

Transcription

1 APPLICATIONS OF CHARGE TRANSFER DEVICES TO COMMUNICATION D. D. Buss W. H. Bailey Texas Instruments Dallas, Texas Texas Instruments Dallas, Texas ABSTRACT Charge transfer devies (CTD's) are analog in operation, and as suh, they are uniquely appliable to many analog signal proessing funtions in the areas of ommuniation, radar, sonar, guidane and ontrol, et. This paper is limited to military ommuniation systems and addresses the following topis: (1) mathed filtering in spread spetrum ommuniation, (2) bandpass and low pass fi 1 tering, (3) Hi 1 bert transform for single sideband modulation, (4) omplex oding for M-ary ommuniation, and (5) adaptive equalization for MODEM's. The CTD transversal filter whih is the fundamental building blok for the above appl iations is desribed and the effet of imperfet harge transfer effiieny on its performane harateristis is determined. Examples of CTD filters are drawn from both harge oupled devies (CCD's) and buket brigade devies (BBD's). However, design and fabriation tehniques are not overed. This paper also summarizes CTD harateristis suh as harge trans-. fer effiieny, dynami range, tap weight error, leakage and bandwidth as they apply to ommuniation filtering appliations. It shows that the ost and power onsumption with CTD filtering an be signifiantly lower than with onventional digital filtering. Spread spetrum systems whih utilize CTD's are ompared with systems whih utilize surfae wave devies and those whih perform the mat~ed filtering digitally. Comparisons are made with respet to performane, ost, power, weight, and size. The effets of harge transfer loss and tap weight error on reeiver sensitivity are alulated, and the ultimate performane limitations of CTD systems are delineated. Data are presented on an inoherent reeiver whih performs mathed filtering at baseband with hirp signals of time bandwidth produt TdW = 1. Measurements of bit error probability show a.5 db sensitivity Joss from minimum bit error probability predited for nonoherent FSK. Results are presented on a 11-stage CCD bandpass filter having Dolph-Chebyshev weighting and the harateristis of this filter are disussed. The remaining topis are overed briefly. I. INTRODUCTION Charge Transfer Devies (CTD's) are analog, sampled data delay Jines, and as suh they are readily appliable to a large number of analog signal proessing funtions. Certain types of sampled data filters whih have, up until now, been implemented digitally,! an be realized in integrated form with CTD's. This paper disusses the advantages and limitations of CTD 1 s for sampled data filtering in military ommuniation sys terns. CTD's inlude two important lasses of devie whih are funtionally very similar: harge oupled devies2 (CCD's), and buket brigade devies3 (BBD's). CCD's are by far the better known, and although they were first announed a sant three years ago, 4 they are already ommerially available. BBD's are older than CCD's and an be fabri- 83

2 ated using onventional MOS proesses. CCD's require non-standard proessing, but they have performane advantages over BBD's and will probably dominate analog signal proessing in the future. The basi building blok whih is used in the appliations overed in this paper is the transversal, non-reursive CTD fi 1 ter. Tehniques for designing and making CTD transversal filters are desribed in the literature5,6,7 and are not disussed here. Instead, this paper deals with those properties of CTD filters that determine their performane in ommuniation filtering appliations. Sine the impulse response of a CTD filter an be hosen arbitrarily, CTD filters an be mathed to arbitrary signaling waveforms in muh the same way that surfae wave devies (swgs) are u~ed to perform mathed filtering. CTD's do not ompete diretly with SWD's for this appliation, however, beause SWD's have higher bandwidth and limited time delay. CTD's on the other hand are apable of proessing signals of up to 1 se in.time duration but are more restrited in bandwidth. The ompetition faed by CTD 1 s for this appliation is onventional digital proessing, and that omparison an be summarized as follows. Compared with digital implementations, CTD filters are less flexible and are limited to time bandwidth (TdW) produts of 13 or less. On the other hand, when CTD's an be used for a partiular mathed filtering funtion, overwhelming advantages in ost, power onsumption, size, weight, and reliability an be realized. Setion II of this paper gives a mathematial treatment of CTD transversal fi 1 ters. Dispersion whih results from imperfet harge transfer is haraterized, and its effet on mathed filter sensitivity is alulated. An example of a CCD mathed filter is given. Setion III disusses the system aspets of CTD spread-spetrum reeivers, and test results on a breadboard system are given. Setion IV disusses bandpass and lowpass filtering, Hilbert transform for single side band modulation, omplex oding for M-ary ommuniation, and adaptive equalization for MODEM's. Setion V onludes the paper with a disussion of the future of CTD's in military spread-spetrum systems. II. CTO TRANSVERSAL FILTERS A blok diagram of a CTD, sampled data, transversal fi 1 ter is given in Fig. 1. It onsists of M delay stages Ok, k = 1, M, together with iruitry for performing the weighted summation of the node voltages vk. Eah delay stage onsists of p transfer eletrodes in a p-phase CTD (e.g. p = 3 for a 3-phase devie). The input to the filter is sampled at the lok frequeny f, and the delay of eah stage is T = 1/f. The filter output is given by M v t (nt ) = :E hk vk(nt) ou k=l M ~ hk v. [(n-k)t 1 k=l In (I) (2) (3) where T = MT is the total time delay of the fil~er, agd h(t) is the impulse response whih has samples hk = h(kt). A. DISPERSION DUE TO CHARGE TRANSFER LOSS ; The filter desribed above has ideal delay stages, bu~ CTD delay 1 ines are not ideal_. The CTO operates by transferring harge from one storage loation to the next, and eah time a transfer is made, a fration a of the harge is lost. This frational loss per transfer a is related to the frational loss per delay stage e by the number of transfers p required to aomplish one stage of delay. e = pa (4) For CCQ4s p = 2, 3, or 4 and typially_ 3 a~ 1. For BBD's p = 2 and a ~ 1 although a~ 1-4 an be ahieved using the tetrode onfiguration.9. The effet of harge transfer loss an most easily be deshibed by defining the z-trans- form of the kt node voltage Vk(z). Using this, imperfet harge transfer an be haraterized by the relation Vk+l (z) = z-l [evk+l (z) + (1-e) Vk(z)] (S) 84

3 whih an be solved to give Vk+l (z) = ( l --~) z-l Vk(z) (6) 1-ez Eq. 6 is the fundamental transfer relation and redues to the ideal delay relation when E:... The relation between v 1 (z) and V. (z) depends upon the atual input iruit. loinhowever, the differenes whih result are insignifiant. For simpliity we will hoose the input tehnique whih results in v 1 (z) = ( 1 - _ E:) &Z -1 z V. (z) 1n Combining eqs. 1, 6, and 7 gives M V t (z) '" }: OU ka) from whih the transfer funtion (...;.._..::_=-:--1 I - e )k hk z -k 1- &Z an be defined. The supersript indiates that the transfer funtion depends upon e. Note that if the ideal transfer funtion HI(z) is determined from eq. 9 by setting &.., I M -k H (z) = }: hk z (1) k=l then the transfer funtion of eq. 9 an be obtained by replaing z in the ideal transfer funtion by Beause in all ases of pratial interest &<<1, eq. 11 an be well approximated by taking (7) (8) (9) (11) ( 12) The above disussion indiates that the dispersion due to harge transfer loss an be alulated by replaing z in the ideal transfer funtion with z'. That is (13) The dispersion due to harge transfer loss.an be viewed in the frequeny domain by using the definition of z -I z = exp[-i2nf/f] (14) together with eq. 13. where The result is HE:(f) ~H 1 (f') (15) ef f' = f + 2 TT {sin(2nf/f)-i[l-os(2nf/f)j} (eq. 16). Eq. 16 states that harge transfer loss introdues a frequeny shift whih is both real and imaginary. At low frequeny the real part dominates as an be seen by expanding eq. 16 in a Taylor series in f/f. For f«f f'~f(l+e:) (17) B. DISPERSION COMPENSATION If the dispersion due to harge transfer ineffiieny is known, the filter an be designed to invert this dispersion. Consider a filter having an infinite number of sampling stages and weighting oeffiients h~, k=l,.,. Its transfer funtion H (k) an be shown by expanding eq. 9 to be He (z) = f: r ~~ h~-. (k:''.j (1- ) k-j] z -k k=l ~= J J J. (eq. 18). By equating the terms in the retangular brakets to the desired weighting oeffiients hk, the desired transfer funtion an be obtained, and the hk whih give rise to the desired transfer funtion are obtained by iterating the relation h' = k k-1 L: j=l (1- ~k (19) Even if the desired impulse response is finite in time (hk = fork> M), the exat solution of eq. 19 requires an infinite number of oeffiients h'. However, in most ases 1 the error whi~ results in trunating the hk series at M terms is negligible. This tehnique for ompensating harge transfer loss has been demonstrated.7 However, the tehnique is limited by the fat that e I l I

4 annot always be predited with suffiient auray, and that it varies with lok frequeny and other operating parameters. C. SENSITIVITY VS. CHARGE TRANSFER LOSS A CTD transversal filter an be used to detet signals in the presene of noise, and for this appliation, the filter impulse response is hosen to be the time inverse of the signaling waveform. Suh a filter is said to be mathed to the signal, and the mathed filtering theorem of statistial ommuniation states that when the noise environment is white and additive the mathed filter provides the optimum output signal-tonoise ratio (SNR).ll Charge transfer loss, however, introdues a slight mismath between the filter and the signaling waveform and results in loss of sensitivity. It will now be shown how harge transfer loss affets the sensitivity of a CTD mathed filter assuming no measures are taken to ompensate for loss. Consider a filter whose weighting oeffiients are hosen to represent an impulse response h(t) whih is mathed to a partiular signal waveform v (t). s h(t) = (1/A)vs (Td - t); (2) where A is the amplitude of the signal waveform and h(t) Is normalized so that its maximum amplitude Is unity. When v Is applied to an ideal the ou~put is given by eq. 2 M v t (nt ).. A :E hk hm- +k ou k=l n and the orrelation peak ourring at tm = MT is M vout(tm) =A 1: h 2 k=l k filter (e:=o), (21) (22) However, harge transfer loss hanges the effetive weighting oeffiients of the filter to new values h: whih an be alulated as in eq. 18 to be and the output signal at the orrelation peak is whih an be written as A f e -1-1 vout(tm) =2TTT H(z) H (z )z dz +f /2 ~ A [ H (f) H *( f 1 ) d f -f /2 (24) (25) (26) In the above equations, H(z) and HE:(z) are the z-transforms of hk and h~ respetively and f 1 is given by eq. 16. The peak output power 2 Sp = vout(tm) (27) an be alaulated from eq. 26 and dereases with inreasing e: beause harge transfer loss introdues an effetive mismath between the filter and the signal waveform. However, harge transfer loss also affets the oufput noise power. If the input noise is assumed to be white and to have single sided spetral density N, the rms noise power is given by Np = : ~~ f He(z) HE:(z- 1 ) z":"l dz (28) N -ff. +f /2 2 (29) provided the CTD mathed fl Iter Is preeeded by an anti-a! lasing filter whih eliminates omponents of noise higher in frequeny than f /2. Under these onditions the output SNR Pout an be evaluated by ombining eqs. 26, 27, and 29. p out s =.J?.. N p. ~- - 86

5 ['J', 2A 2 -f /2!" -- N -f /2 (f) H* (f 1 ) IH(f '> 1 2 df df 2 (3) Contat an be made with the familiar equal! tion for the output SNR of a mathed filtef by letting e -. In this ase 2E pout... ~ (31) where the signal energy E is given by s +f /2 Es=A2 ( \H(f)\ 2 df (32) -f~2 A very useful result an be obtained by expanding eq. 3 in a Taylor series in e. If p is written as out 2Es Pout "" -N- - Qle - 2 f3e (33) then it an. be shown that Ql =. Even though the output signal power s dereases linearly with due to signal mi~math, the output noise power N also dereases proportionally, so that on~y terms seond order and higher in e ontribute to sensitivity loss. This is extremely fortuitous beause It allows one to design filters having fairly large dispersion (large Me produt) without seriously degrading sensitivity. The above analysis is illustrated with alulations made on a 1-bit pn filter whose pseudorandom ode is given in Table 1. The output peak signal power S and the output rms noise power Np are nor~alized to their respetive Input values (A2 and N f /2) and plotted in Fig. 2. Both quantitigsderease linearly with e but the output SNR P t dereases muh more slowly. ou Fig. 2 an be used to determine an upper 1 i mit on the dispersion whih an be tolerated in a mathed filter. If it is arbitrarily speified that harge transfer loss degrade the sensitivity by no more than 9% (.5 db), this requires the Me produt to be less than r. 6. The Nyquist sampling theorem (W < f /2) together with the above limitation onme produt (Me< r) leads to the following limitation on time bandwidth produt TdW: T W ~ M/2 ~ _[_ (34) d 2e Using e = 3 x 1-4 this gives TdW$ 13. D. OTHER LIMITATIONS In addition to the loss whih results when harge transfers from one storage loation to another, CTD's lose harge due to leakage. This limits the total time delay whih an be ahieved to 1 se.or less. For all CTD's harge transfer loss inreases severely at high frequenies, and although CCD's may eventually operate at hundreds of MHz, urrent tehnology 1 imits CTD filters to 1 MHz or less. These limitations, together with the TdW limitation of the previous setion, are illustrated in Fig. 3. Tap weight error is the 1imiting 1 fator in some CTD filtering appliations. However, for mathed filtering appliations, the additional "noise" introd~ed by random tap weight is negligible. E. EXAMPLE The operation of a CCD filter is shown in Fig. 4. This devie was designed using the eletrode weighting tehniqueland the hk oeffiients are given in Table 1. No intereletrode gaps are present in the photomirograph at the top of the figure beause lhkl = 1. The filter impulse response is shown at the enter of the figure and the orrelation response is shown at the bottom. Note the orrelation peak in the output. The devie shown had Me~ o. 1, so the output waveform is essentially ideal. II I. SPREAD SPECTRUM COMMUNICATION USING CTD'S As indiated in the previous setion, CTD's are limited in frequeny to around 1 MHz. Therefore, filtering in a CTD system must be performed,at baseband in distintion to SWD systems where the filtering is performed at RF. 87

6 Figures 5 and 6 illustrate the differene between RF and baseband proessing. Assume the signaling waveform Is of the form v. (t) =A h. (t) os(w t +If>) I I i =, I (35) where If> is the unknown phase and where h and h1 represent the two binary waveform~ being transmitted. Figure 5 shows the RF proessor in whih the inoming signal is first filtered in a filter mathed to the RF waveform and then envelope deteted. Envelope detetion is indiated here as a mixing operation followed by squaring and adding operations. In the baseband system of Fig. 6, the mixing and filtering operations are interhanged. The inoming RF signal is first mixed to baseband, and then filtered in a filter mathed to the baseband signal. Both systems an be shown to have idential performane, and their bit error probability is equivalent to nonoherent FSKI3, Equation 35 is not the most general type of narrow band signaling waveform whih an be employed. The most general form Is v. (t) =A [h. (t) os(w t +If>) I I =, I (36) This generalized waveform ompliates the baseband proessor of Fig. 6. Four filters are required mathed to h, h 1, g, and g 1. The disussion will now be speialized to hirp (linear FM) signaling systems. Chirp is partiularly attrative beause It is minimally sensitive to error in the loal osillator frequeny, and this is expeted to be a major problem in CTD spread spetrum sys terns. The RF hirp signal an be written as: v. (t) I A os ( w t ± 1J. t 2 + If>) -T /2 < t < +T /2. d d (37) where i = I orresponds to an up-hirp (+) and i = orresponds to a down-hirp (-). For onveniene the time interval is taken to be symmetri about t =. The RF signal hirps. through a bandwidth entered about f w =- 2TT (38) The hirp signal of eq. 37 is of the form of eq. 36 as an be seen by expanding eq. 37. In this ase 2 h = h 1 = os 1-Lt Td Td -2<t<r The use of hirp also simplifies reeiver design somewhat beause now two pairs of filters are required mathed to the two signals of eq. 39. (39) A baseband system whih an be used to detet binary hirp is shown in Fig. 7. The filters marked SIN and COS are mathed to the signals of eq. 39. In a baseband system suh as that of Fig. 7, the baseband signals hirp from -W/2 through zero to +W/2 or vie versa, and eah filter has a bandwidth of only W/2. Therefore to reover all the information in the signal two filters are required, and their outputs must be added oherently as shown. The hirp through zero sheme has the further advantage of halving the bandwidth requirement on eah filter. The I - q deision Is made by omparing the up-hirp output with the down-hirp output at the time of the orrelation peak. The system of Fig. 7 was implemented using 2-stage BBD filters. The signaling waveforms have TdW = 1 for an overall proessing gain of 23 db. To eliminate aliasing, the filters were designed to sample the impulse responses of eq. 39 st 2W, i.e. twie the Nyquist rate for the highest frequeny. The responses of these filters to a negative impulse are shown in Fig. 8. The derease in the amplitude of the impulse response~ results from harge transfer loss whih for these devies was a= Jo-3. The sensitivity Joss whih this auses is al~ ulated to be less than.i db. The orrelation responses whih result from a oherent hirp signal are also shown in Fig. 8. Fig. 9 shows the outputs of the up-hirp and down-hirp hannels when the input signal is nonoherent, and hirps alternately up and down. Even when the input signal Is 88

7 masked by noise the o.rrelatlon peaks in the output are unambl~uous. The bit error probabi.lity P for a nonoherent FSK reeiver is given 5yl3 where P.. 1/2 exp(-y/2) e v.. Es I N (4) (41) E is the signal energy, and N is the single sided spetral density of he noise. The measured bit error probability is ompared in Fig. 1 with eq; 4 and shows that the detetor sensitivity is within.5 db of the theoretial limit. The measured dynami range of the filters themselves is in exess of 75 db, and they have been tested from -6 C to +8 C. Devie performane improves at low temperatu~e, but at higher temperature leakage urrent 1n both CCD's and BBD's limits low frequeny (long Td) operation. IV. OTHER COM~UNICATION APPLICATIONS Within the onstraints of (27) - (29), the appliability of CTD's to sampled data filtering problems is limited only by the designer's imagination. Some of the more ommonly ourring appliations are disussed in this setion. A. BANDPASS FILTERING A CTD transversal filter an be used to implement a bandpass filter by seleting the impulse response of the filter to be the inverse transform of the frequeny harateristi.12 The measured frequeny response of suh a filter implemented with CCD's is given in Fig. 11. This filter was designed using Dolph-Chebyshev weighting to ahieve 29 db rejetion and a 3 db bandwidth of 4% of the enter frequeny. The enter of the passband ours at f /4 and the filter is tunable by varying f~. Using this approah, it is diffiult to ahieve filters of high Q beause of the finite time duration of the impulse response of a transversal filter, and beause tap weight error limits the out-of-band rejetion. Reursive filters are potentially useful for ahieving high Q as has been demonstrated using BBD's14 with off-hip feedbak. However, until tehniques are devel- oped for integrating the req~ired feedbak, CTD reursive filters will not be generally appliable. CTD transversal filters are most useful when the magnitude and phase of the frequeny response must be aurately determined. They an also be used when the signal level is low (down to 5 ~volt) and have large dynami range. B. HILBERT TRANSFORM A Hilbert Transform onsists of a onvolution with t-1 and an be implemented with a CTD transversal filter having weighting oeffiients k _ (M+I) -2- k = I, H H even (42) Suh a filter an be used to generate single sideband signals 1 and has a wide potential appliation in ommuniation equipment. C. COMPLEX CODING CTD filters an be used to generate as well as to detet arbitrary waveforms. This presents the possibility of hoosing M waveforms to represent k = log 2 M bits of information. This type of oding is urrently employed in MODE~'s, but the hoie of waveform is limited by the available equipment, and the waveforms atually used are not orthogonal or optimized to the transmission medium. CTD~s offer a great deal of flexibility in seletion of waveform. In addition, the relative ease with whih CTD filters an be mathed to arbitrary waveforms may make feasible the use of M-ary (instead of binary) ommuniation where it has not previously been pratial. D. ADAPTIVE FILTERING Due to the hanging harateristis of a telephone hannel, adaptive equalization is required in a MODEM. This requires a variable tap weight onvolution filter (VTWCF), and although all CTD onvolution filters reported to date have fixed weighting oeffiients, a CTD V~1CF is feasible. Suh a filter would be extremely useful not only in HODE~'s but in other adaptive filtering appliations suh as voie reognition, adaptive beam forming, remote intrusion dete- 89

8 tion, et. V. CONCLUSIONS The largest impat whih CTD's will make on military ommuniation systems is ost redution of systems whih are produed in suffiiently high volume to offset the development ost of a ustom CTD filter. The ost omparison between CTD and digital implementations is extremely diffiult beause it must be made for eah system. However, for i 1 lustrative purposes we will give a simple example. It is required to implement a 1 point onvolution filter with a dynami range whih requires 8-bit digital logi. An analog signal is to be sampled at 1 khz, filtered and presented at the output In analog form. When this filtering operation is implemented using TTL the ost and power onsumption are determined primarily by the 8-bit AID onverter ($2, SW) and the digital filter (approximately 4 TTL networks osting $5 apiee and dissipating 1 mw apiee). In evaluating the ost of the CTD implementation the important fator of part volume enters. The digital filter an be implemented with standard atalog items whereas the CTD is a ustom IC, and as suh its ost is strongly dependent upon the number of parts required. A reasonable approah to alulating the prie of a CTD ustom hip is to assume 8 k$ development ost plus $5 per opy. Using this, the prie per opy p would be 4 p = $5 + $8 ~ 1 (43) u where N is the total number of units required.u If N = 8, then p = $6 for the CTD. In ~ddition to the CTD, iruitry is required for lok drivers, output amplifiers and output sample-and-hold. The total ost for these parts is less than $5, and the power dissipation is on the order of 6 rrw. A summary of this omparison is given below: Cost Power Digital $4 4SW CTD $56 6rrW This omparison is an oversimplifiation of real systems, and it is heavily weighted in favor of CTD's beause it is based upon a simple onvolution filtering operation for whih CTD's are ideally suited. However, it illustrates the saving in ost and power whih is possible with CTD's. CTD's have potential advantages in small size and low weight whih result from the ompatness of the filter IC and potential advantages in inreased reliability whih result from a redution in the number of pakage interonnets required, Communiation is an important field whih appears to be ideally suited to apitalize on the advantages of CTD filtering. Progress has been made in developing CTD tehnology for these appliations, and ontinued development at this time is amply justified. ACKNOWLEDGEMENTS Most of the work presented in this paper was sponsored by Rome Air Development Center under ontrat #F C-27 and direted by Charles N. Meyer of RADC. Development of the CCD filter of Fig. 4 was sponsored by the ~aval Eletronis Command under ontrat #N39-73-C-13 and direted by Dr. David F. Barbe of N.R.L. Development of the CCD bandpass fi Iter disussed in Se. I V-A was sponsored by the U. S. Army Eletronis Command under ontrat #DAAB7-73-C-266, and direted by Ted J. Lukaszek of Ft. Monmouth. REFERENCES 1. B. Gold and C. M. Rader, Digital Proessing of Signals, MGraw-Hill, W. S. Boyle and G. E. Smith, "Charge Coupled Semiondutor Devies," Bell Syst. Teh. J., 49, pp , April F. L. J. Sangster, "Integrated MOS and Bipolar Analog Delay Lines Using Buket Brigade Capaitor Storage," in 197 IEEE Sol id-s tate Ciruits Conf., Dig. Teh. Papers, pp , Fairhild's CCD 11 is a 5-element eo 1 i near image sense r. 5. D.. Buss, W. H. Bailey and D. R. Collins, "Mathed Fi 1 tering Using Tapped Buket-Brigade Delay Lines," Ele. Lett. 8, pp , 4 Jan

9 6. D. R. Collins, W. H. Bailey, W. M. Gosney and D. D. Buss, "Charge-Coupled Devie Analog Mathed Filters," Ele. Lett. 8 pp , June 29, D. D. Buss, D. R. Collins, If. H. Bailey and C. R. Reeves, "Transversal Filtering Using Charge Trasnfer Devies," IEEE J. Solid-State Ciruits, SC-8, pp , Apri D. T. Bell, Jr., J. D. Holmes, and R. \1. Ridings, "Appliations of Aousti Surfae Wave Tehnology to Spread Spetrum Communiations" (Invited Paper) IEEE Tran. Mirowave Theory Teh., MTT-21, pp , Apri L. Boonstra and F. L. J. Stangster, "Progress on Buket-Brigade Charge Transfer Devies" in 1972 IEEE Solid State Ciruits Conf., Dig. Teh. Papers, pp , D. D. Buss, W. H. Bailey and D. R. Collins, "Analysis and Appliations of Analog CCD Ciruits," Proeedings of 1973 International Symposium on Ciruit Theory, Apri , 1973, Toronto, pp M. Shwartz, W. R. Bennett and S. Stein, Communiation Systems and Tehniques, MGraw-Hi 11, 196h, pp D. D. Buss, D. R. Reeves, W. H. Bailey and D. R, Collins, "Charge Transfer Devies in Frequeny Filtering," Proeedings of the 26th Annual Symposium on Freque~y Control, June 6-8, 1972, Atlanti City, N.J., pp M. Shwartz, 11. R. Bennett and S. Stein, Op. Cit., pp , D. A. Smith,. M. Pukette, and \.J. J. Butler, "Ative Bandpass Filtering with Buket-Brigade Delay Lines," IEEE J. Solid-State Ciruits, SC-7, pp , Otober FIGURES Fig. 2 N :r ;;= Me Signal power, noise power, and SNR at the output of the 1-bit filter desribed in the text. These results are normalized to their respetive quantities at the input to the filter and plotted as a funtion of the loss parameter Me. Fig. 1 Blok diagram of a transversal fi 1- ter onsisting of M delay stages Dk and H weighting oeffiients hk. Fig. 3 The approximate limitations on the time duration (Td) and bandwidth (W) of signals wfiih an be proessed using CTD's. T ~ 1 se, W Co; 1 HHz, T W t; lo3? d 91

10 C :;:o FN SEOUt~C MATCHED FILTER Fig. 6 Blok diagram of a binary baseband proessor. IMPU1..S RESPONSE INPUT CCRR[LATIO~ RE.S?OtiS( Fig. 4 (above) Operation of a 1-bit pn sequene eo filter mathed to a pseudorandom ode. Top: Photomirograph of the first few stages of the filter. Center: Response of the filter to a negative impulse. Bottom: Correlation response of the filter. Note the orrelation peak in the output waveform. Fig. 7 A reeiver for binary hirp waveforms. The boxes labeled COS and SIN are BBD filters whose harateristis are shown in Fig. 3. IMPULSE RESPONSE h(tlosut 2 -f<t<~ CORRELATION RESPONSE IMPULSE RESPONSE CORRELATION RESPONSE Fig. 5 Blok diagram of a binary RF proessor. Fig. 8 The impulse response and orrelation response of the two filters required for the system of Fig. 7. These filters are eah 2 stages long and are both integrated in a single BBD IC. 92

11 UP-CK!RP OUTPUT DfiWII CHRIP OUTPIJT INPIJT 'IJ?-CHRIP OUTi'UT DOWN-CHIRP OUTPUT ORTHOGONAL MATC!-i FILTER CORRELATION RESPONSE WITHOUT NOISE Figure 9. The outputs of the reeiver.of Fig. 7 when nonoherent up-hirp and down-hirp signals are sequentially applied to the input. Even when the input is obsured by noise, the output is unambiguous. -1 "' -2 "' ~ z ~ -JO,.~. A ~ I«~ tHl 1\ ~~~ FREQUENCY ~HZI ORTHOGONAL ~ATCHEO!='ILlER CORRELATION RESPONSE WITH 8 lnput SIGNAL-TO-NOISE. RATIO LO Fig. 11 The measured frequeny response of a 11-stage Dolph-Chebyshev CCD bandpass filter. The filter was designed to have 29 db outof-band rejetion and tunable passband at.25 f THEORETIC'! NONCOHERENT FSK Code for the 1-Bit E'1 Sequ~ne F'ilter The til'lle signal onsists of Hoe ~le;r-e:'lts of this t<1ble in the order given, The fll ter itself is oded wit~'. these values in reverse o,.der. o DOWN CHIRP UP'CHIRP CHIRP BW 44kHz T?.27 MS NOISE BW 44 khz ) 31-4,., - so Sl o JO ll INPUT SNR, <db! Fig. 1 Measured bit error probability P of the reeiver of Fig. 7. The e overall sensitivity was found to be.5 db less than the ideal sensitivity for nonoherent FSK. REVERSE SIDE BLANK 93