You are on page 1of 15

24

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7, NO. 1 , JANUARY 1989

Real-Time Long Range Complementary Correlation Optical Time Domain Reflectometer

Abstract-The performance of an optical time domain reflectometer (OTDR) is significantly improved using a new correlation technique employing codes with complementary autocorrelation properties. The theoretical foundations of the new method are presented along with experimental results exhibiting the best one-way range reported to date for a practical long-haul long wavelength OTDR system.

namic range of an OTDR and significantly reduce the required measurement time, without sacrificing resolution. The resulting performance is the best yet reported for a practical system.
11. BACKGROUND A. Sensitivity-Resolution Trade-off A fundamental limitation to any conventional OTDR measurement [2] is the trade-off between response resolution and signal-to-noise ratio (SNR). The response resolution of a measurement is the minimum separation between two faults that allows these faults to be clearly distinguished. Increasing the SNR of an OTDR measurement results in increased dynamic range in a given measurement time, or, alternatively, a reduction in the number of averages required to obtain a given result. The backscatter impulse responsef( t ) is defined as the backscattered power signal in response to a unit energy delta function injected signal. For an ideal fiber with constant propagation loss along its length, f( t ) is given by:

I. INTRODUCTION URING THE PAST several years, the task of the optical time domain reflectometer (OTDR) measurement [l], [2] has become more difficult. As the users of optical fiber have migrated to longer transmission wavelengths because of lower loss, and as higher quality fiber has become available, there is literally less backscattered light to be measured. The dominant use of single-mode fiber results in a reduction in the backscatter coefficient and in the input/output coupling of the probe signal with respect to multimode fiber OTDRs. At the same time, in addition to the high sensitivity needed to measure extremely low levels of backscattered light, the long lengths of the fiber systems that are being installed require that OTDRs also be able to measure a very large dynamic range of signals. Yet, despite their wide usage during the past several years, surprisingly few fundamental changes in the design of practical OTDRs have been seen. By practical OTDRs we refer to measurement systems using semiconductor sources that can be potentially implemented in rugged field-portable instruments. While incremental changes such as the use of higher power lasers and the parallel averaging qf all data points on a single probe shot have provided some improvement, they hqve been unable to remedy a limitation that is fundamental to all conventional OTDR designs: the trade-off between signal-to-noise ratio (which determines dynamic range apd measurement time) and response resolution. This paper describes a new approach to optical time domain reflectometry that applies spread-spectrum techniques, such as those used in radar, to increase the dy-

f(t)

icrv,SPine

(1)

where cr is the attenuation coefficient of the fiber, ug is the group velocity of the probe pulse, S is the backward capture coefficient [3], and Pinis the optical power injected into the fiber. In any OTDR measurement, the received optical power P ( t ) can be expressed as the convolution of p ( t ) , the optical power signal that probes the fiber and the backscatter impulse response f( t ) which is the signal to be measured:
= A t ) 8f G > . (2) The detected current is related to the return intensity by a factor q which is the product of the coupling losses, and the detector responsivity :

i(t) = qP(t).

(3)

Manuscript received October 19, 1987; revised December 21, 1987. M. Nazarathy, S . A. Newton, 8.P. Giffard, D. S . Moberly, W. R. Trutna, and S. Foster are with Hewlett-Packard Laboratories, Palo Alto, CA 94304. F. Sischka is with Hewlett-Packard GmbH, 7030 Boblingen, West Germany. IEEE Log Number 8821129.

The output voltage is given by the convolution ( 8 )of the current with the transimpedance impulse response:
u(t) = r(t) 8i(t).

(4a)

Combining the last three equations, one obtains

4t)=

r r ( 4 6 P ( t >@ f ( t > .

(4b)

0733-8724/89/0100-0024$01.OO 0 1989 IEEE

NAZARATHY et a l . : LONG RANGE COMPLEMENTARY CORRELATION OTDR

25

The response resolution of the measurement is therefore degraded according to the duration of the receiver response as well as that of the probe signal. In a conventional single-pulse OTDR, the probe signal is simply a square pulse having duration AT. If A T = 1 p s , for example, the response can be no better than approximately 100 m, even with an ideal receiver. Strictly from the point of view of response resolution, it is desirable that the probe pulsewidth be as small as possible. On the other hand, the strength of the received signal is proportional to the energy of the probe signal, which is the product of the peak power and the probe pulsewidth. In a conventional single-pulse OTDR, the initial signal strength resulting from a short single probe shot of width AT is approximately given by
P,,it

the measurement range and measurement time on the SNR. The trade-off between the SNR and the response resolution is therefore an important and fundamental limitation on the overall performance of a conventional single-pulse OTDR.

B. Spread-Spectrum Techniques Spread-spectrum techniques such as correlation [2] offer the possibility of providing measurements with improved SNR without sacrificing response resolution. Such techniques are commonly used in radar [7] and other peakpower limited systems where increases in the transmitted energy would otherwise result in degraded resolution. One way of applying spread-spectrum techniques to OTDR measurements is to correlate ( * ) the detected signal v ( t ) with the probe signal p (t):
P(t) * 4 t ) = P ( d

~cxv,SP~,AT.

This readily follows by performing the convolution of a short rectangular pulse with the impulse response (1). The initial backscatter level can be maximized only by increasing either the peak power or the duration of the probe pulse. Unfortunately, the use of very high-power sources [4], [5] is precluded in a practical system due to portability and durability requirements, which point to the use of semiconductor laser sources. The use of special high peakpower semiconductor lasers is also problematic in a practical system, due to their high cost, limited lifetimes, and electromagnetic interference of the high drive currents with the sensitive receivers. Another area of improvement is the design of low-noise receivers [6]. Unfortunately, such low-noise designs require photodiode cooling and their noise performance rapidly deteriorates for bandwidths over 1 MHz, due to the onset of noise sources with quadratic frequency dependence of their noise spectral density. Also, trade-offs exist between receiver sensitivity and overload recovery for such receivers. Once the peak pulse power has been increased to its maximum practical value, and the best available receiver has been incorporated, the probe energy and therefore the return signal can be further increased only by making the probe pulse longer, which in turn degrades the response resolution of the measurement. The trade-off between signal strength and resolution in a single-pulse OTDR manifests itself as a limitation on the dynamic range or, alternatively, on the averaging time of the measurement. The SNR (in decibels) of a measurement at a range z can be simply expressed by an OTDR-makers formula:

* [ P W @ V ( t >@ff(t)l = [At)* P ( O ] @ [ V W 60 f ( t > I .

(6)

To the extent that the autocorrelation of the probe signal approximates a delta function, the fiber backscattering response f ( t ) can be accurately recovered, subject (as always) to the response of the receiver: P(t>*

4 4 - W ) @ [rlr(t) @ f W l
(7)

where ( ) denotes proportionality. In this case, it is the duration of the autocorrelation of the probe signal that determines the response resolution rather than the duration of the probe signal itself, which may be long and energetic. The idea of using correlation in OTDR measurements is not new. A number of proposals and experimental demonstrations have been published previously [9]-[ 151. In each of these experiments, pseudorandom codes were used to probe the fiber under test. More recently, the use of Barker codes in connection with a coherent OTDR was demonstrated [16]. This previous body of work may be divided between methods which use periodic codes [9]1131 versus aperiodic codes [14]-[16]. Although there exist periodic codes with good autocorrelation properties, they are generally unsuitable for practical systems as they pose a serious dynamic range problem due to possible saturation of the front end electronics by Rayleigh scattering from the early part of the fiber, and by strong Fresnel reflections. Aperiodic codes with extents shorter than the repetition SNR = Pldt - ~ C X - P N E + 1.5Nmt Z (5) period of the OTDR measurement allow in principle a sowhere PNE the receiver noise equivalent power in dBm, lution to the dynamic range problem. However, aperiodic is N is the number of probe shots, and No,, is N expressed unipolar sequences with zero-sidelobes are not known to in octaves, i.e., No,, = logz N. For example, each time exist, whereas aperiodic codes with good autorcorrelation the number of averages is doubled, i.e., No,, increases by properties exist, but are hard to find. Research first fo1 , the SNR improves by or 1.5 dB. (The use of 10 cused on identifying unipolar aperiodic sequences with loglo P optical decibels is assumed throughout the pa- low autocorrelation sidelobes and demonstrating an OTDR system based on such sequences [ 141. Unfortunately, per ). The OTDR-makers formula clarifies the dependence of stringent requirements are posed for the mainlobe to side-

= V ( t ) @f(d.

A,

26

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7, NO. 1 , JANUARY 1989

lobe ratios, in order to avoid distortion in the neighborhood of large reflections and discontinuities which may exceed 30-dB optical in magnitude. Another approach [15] pursued the synthesis of zero sidelobes by summing autocorrelations and cross correlations of binary counting sequences or subsets of sequences such as cyclic shifts of M-sequences or quadratic residue codes. For long code lengths, the size of such subsets might become prohibitive, and the correlation gain is offset by the processing time spent in computing a large number of correlations. Although, to the best of our knowledge, the code hopping complementary technique [ 151 was never experimentally demonstrated, that approach nevertheless was the closest predecessor of the new technique presented in this paper in terms of its mathematical equivalence. In this paper we present and experimentally demonstrate a complementary autocorrelation technique based on complementary pairs of Golay codes, which leads to practical implementations of OTDR systems with improved relative performance with respect to conventional, single pulse techniques. The practical success of the new technique stems from three elements: 1) The method uses aperiodic codes to solve the dynamic range problem characteristic of correlation techniques which use periodic codes. Aperiodic codes are well suited to adaptive signal processing algorithms, which adjust the signal levels to saturation conditions which may be present within the measurement window. 2 ) Although complementary code algorithms using bipolar codes are known, they are inappropriate for direct detection OTDRs which require unipolar codes. The new algorithm presented here theoretically results in zero measurement sidelobes. Small residual sidelobe levels are still observed in practice due to small nonlinearities and other imperfections. For the experimental system demonstrated here the worst case mainlobe to sidelobe ratio was in excess of 40 optical decibels, compatible with stringent accuracy specifications for OTDR measurements. 3) The signal processing algorithm is implemented by performing a large number of parallel averages, signed additions of averaged waveforms, and a pair of digital correlations. Since only a pair of codes is used, rather than a large number of codes, the correlation-time overhead is kept low. Thus, the use of correlation yields a substantial net SNR gain. This point is evidenced by direct experimental comparisons of single pulse measurements and coded measurements.
In fact, a footnote in [15] states, In this case it is probably better to

We start the description of the new method by introducing the Golay complementary codes in the next section.
111. COMPLEMENTARY CODES The new method described here realizes the full advantage of correlation by probing and correlating with pairs of probe signals that have complementry autocorrelation properties. These probe signals are the complementary codes [17], which were first introduced by M. J. E. Golay in the late 1940s [ 181 as a method of improving the performance of multislit spectrometers. A complementary code pair, otherwise known as a Golay sequence pair, is defined by the following property: Dejinition [17]: A pair Ak, Bk of two of L-element sequences is said to be complementary if the sum of the autocorrelations of the two sequences is zero for all nonzero shifts. Thus, the sum of the autocorrelations of the two member sequences is a (discrete-time) delta function:

Ak where

* Ak

B k

* Bk

= 2L6k

(8)

(9) and the discrete-time cross correlation of two sequences and Y k :


L- 1

Xk

xk * Y k =

m=O

k+L-1

xmYm+k =

m=k

xm-kYm

(lo)

use complementary series as described by Golay. Unfortunately, the actual derivation of complementary Golay code algorithms for unipolar codes was not further pursued in that publication or elsewhere to the best of our knowledge. The concept of bias cancellation presented in this paper is the key to carrying the statement in the footnote of [15] to actual implementation.

for a correlation kernel xk which is finite with nonzero values between k = 0 and k = L - 1. The unique autocorrelation properties of the Golay codes are shown graphically in Fig. 1. The upper two plots (Fig. l(a) and (b)) show the individual autocorrelations of each one of a 64-bit complementary code pair. Each of the individual codes consists of a pattern of +ls. The value of each of the autocorrelation peaks is equal to the number of bits in the individual code. Each of the individual autocorrelations also exhibits sidelobes that are up to 10 percent of the peak height. However, when the autocorrelations are added together, the peaks add together to a value of 2L and the sidelobes cancel exactly. It is this contribution of all of the bits to the autocorrelation peak along with the complete cancellation of the sidelobes that allows the correlation technique to work in practice. In designing a practical system of this kind, it is essential to work with an autocorrelation function that is perfect in principle, since finite sidelobes will always appear in a real, nonideal system. Using complementary codes, the sidelobes in a real system can be low enough so that the full advantage of correlation can be realized. A number of iterative constructions for complementary code pairs have been derived by Golay [171. One of these procedures, known as appending is applied to an

NAZARATHY et al.: LONG RANGE COMPLEMENTARY CORRELATION OTDR

i;&i
20 10
50

27

0 -10 -20 -70 -60 -50 -40 -30 -20 -10

0 tl..

10 20

30 40 50 60 70

(a)

io1

1
0
t1.e

OtO

i
50

-70 do -50 -40 -30 -20 -10

10 20 30 40

60 70

(b)
SUM OF AUTOCORRELATIONS

:6 0 .

70.

Y).

40.

3Q.
20. 10 .

0.
-70 -60 - -40 -24 -20 -IO 9

0 tl..

10 20 34 40 9 60 70

(C) Fig. 1. Complementaly code autocorrelations: (a) and (b) are individual autocorrelations of each code of a @-bit Golay code pair, (c) is the sum of the autocorrelations. Discrete-time points are connected by straight segments in the computer graphics so that they are more visible.

L-element code pair yielding a 2L-element code pair:

where denotes the complement of B, obtained by swapping 1s and -ls, and 1 denotes the concatenation of sequences, e.g., A 1 B is obtained by appending the elements of code B to the right of the elements of code A. Starting with the one element Golay pair the Golay codes of length 2, 4, 8 are readily derived:

signal processing sequence is required. The problem is that the complementary codes are bipolar, and since the OTDR makes use of square-law detection, there is no way to probe the fiber with negative signals. As a prerequisite to the description of our solution in Section VI, it is convenient to first introduce some signal processing principles of system ident8cation using bipolar complementary codes. The term system identification [19] refers to the determination of the impulse response of a linear time invariant system. The obvious method of system identification is to excite the system with a short pulse which approximates a &function. The output of the system becomes an estimate of the actual system impulse response. This estimate is degraded by the finite width of the input impulse by the presence of noise. For a system using a receiver matched to the injected pulse the resolution is roughly the inverse of the receiver (and probe pulse) bandwidth and the sensitivity is proportional to the probe pulse energy. Since the resolution degrades linearly in probe pulse-width, a direct trade-off is set between sensitivity and resolution as already discussed in Section I1 in the OTDR context. Correlation system identification methods have been used in acoustics [20]-[23] to improve sensitivity for the same resolution, making use of codes with low autocorrelation sidelobes (ideally zero) as probe inputs. In the following we describe a simplified discrete time version of a system identification method for the retrieval of the discrete-time samples hk of the analog impulse response, based on bipolar complementary sequences. The following two mathematical properties of discrete time analysis will be often invoked: 1) The output xk of a discrete linear time invariant system is given by the convolution of the input Pk with the impulse response hk:
m

xk = hk @ Pk =

-CO

2 hkPn-k.

(13)

2) Convolution and correlation are distributive and associative. Thus, for any three sequences A , B, C:

[;] [;; -;I


--t

l, 1, --*[I7 1 -1, 1, -1, -1,

-1) 1 1, -1, 1, -1

C*(B@A)=(C*B)@A

(144 (14b) (14c)

C*(B + A ) = C*B

+C*A

1, 1, 1 -1, 1, 1, 1 -1,

C @ (B + A)

C@B

+ C @ A.

(Remember however that convolution is commutative whereas correlation is not). Let Pk be a probe code, with autocorrelation
Pk

This procedure may be continued on to generate sequences of lengths which are powers of two. Notice that all sequences generated this way are bipolar, with f1-valued elements.

* Pk

L6k

ck

(15)

SYSTEM IDENTIFICATION In order to realize the full correlation gain offered by the use of complementary codes as probe signals, a novel
IV.

where co = 0 and c for k # 0 represents the sequence k sidelobes. The output xk of the system under test is further digitally correlated with the probe code yielding an estimate Y k for the system impulse response:
Y k = x k * P k = [ h ~ ~ P ~ l * P k = hPkk ~P k ] (16) [ *

28

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7. NO. I , JANUARY 1989


A

where use was made of (14a). Using (15), this estimate is given by

q T z x
h ,

2Lhk
E 8 em
U"

er e*

hk

0,
XI'

Neglecting the sidelobes ck, it is apparent that the signal is a factor of L stronger with respect to that obtained with a direct system identification scheme which injects a delta function probe signal. However noise also builds up in the correlation operation. Assume that noise is white, i.e., uncorrelated from sample to sample. When the probe signal is bipolar, +l-valued, the correlation reduces to signed addition of L samples, thus the noise contributions add on a root mean squared basis, i.e., noise builds up by The overall SNR improvement with rea factor of spect to a direct impulsive probe scheme is

Y"

Fig. 2. Bipolar system identification procedure. The two codes are injected into the system, outputs are correlated with the respective codes, and the results are summed.

in the sequence:

A.

-=

Jz. Ji

(18)

= 2L6k

8 hk

= 2Lhk.

(21)

The effect of the sidelobes is to introduce a distortion term hk 8 ck in the estimate of the impulse response. For slowly varying impulse responses this distortion will be small since generally ck contains high frequency variations, however if hk contains discontinuities and impulses as in optical time domain reflectometry the effects of this sidelobe distortion will be appreciable. This distortion would disappear for a code Pk which has zero autocorrelation sidelobes, however, such codes are not known to exist. An alternative system identification method, previously used in acoustic problems [22], [23] makes use of a complementary pair of probe sequences, each of which does not exhibit zero sidelobes separately, but the sum of autocorrelations of which is a delta function. The basic system identification procedure (Fig. 2) consists of the following steps: 1) Probe the system under test with complementary codes Ak and Bk and record the respective outputs x f ,
X:

For OTDR, one is particularly interested in measuring an impulse response which is essentially exponential, but may also contain step discontinuities (splices) and spikes (discrete reflections), superposed on the exponential decay. The results of a computer simulation, are shown in Fig. 3 . The partial outputs y f , y f retrieve distorted versions of the impulse response under test, the distortion being attributable to the presence of the autocorrelation sidelobes. However, upon addition of the two correlation outputs , these distortions cancel out, yielding perfect reconstruction of the original impulse response. Notice that the reconstructed response is 2L times larger than the response hk due to a digital impulse. This alone seems to indicate an improvement of the SNR by a factor 2L, however the noise propagated through the system also increases by a factor of 6 L (it builds up by a factor of & for each correlation and by a factor of h when the two correlation results are added.) The net overall improvement in two measurement cycles is then:

and
Xf

=B k

@ hk.

( 19b)

2, these Outputs with the respective probe sequences, forming the partial results y: and y f :
y: = Ak and yf = B k

However, during the two cycles it took to transmit, detect, and process a pair of codes, one might have transmitted a pair of single pulses and averaged over the resulting two responses, improving the SNR by a factor of &. Thus, the relative SNR improvement ratio with respect to the conventional single pulse method is actually
-=

* Xf * XE.

(19c)

d2L

Jz Jz.

This complementary method of linear system identification is applicable whenever it is possible to transmit bi3 ) Sum these partial results to yield an estimate of the polar signals with positive as well as negative content such system impulse response as in acoustics, or in coherent OTDR. However, for an incoherent OTDR, the on-off keying of optical power only Yk = Y f + Y f (20) provides unipolar (nonnegative) signals, curtailing direct Ideally, the output is equal to the impulse response under transmission of bipolar content. A novel algorithm detest amplified by the factor 2L, twice the number of bits scribed in Section VI applies unipolar complementary (194

NAZARATHY et al.: LONG RANGE COMPLEMENTARY CORRELATION OTDR

29

Fig. 3. Computer simulation for system identification. (a) Correlation Output d . @) Correlation output Y f . (c) Reconstructed response Y k = y: + YP.

L
fo $0)= f 0)
f(0)
= +U,.

Fig. 4. (a) OTDR signal path from the input of the laser driver to the output of the analog to digital converter. (b) Equivalent system representing the signal path as a single discrete-time convolution.

code intensity modulated signals for backscatter impulse response identification. In preparation, a discrete time interpretation of the operation of an OTDR is first derived in the next section. V. EQUIVALENT DIGITAL SYSTEM In this section we follow the signal path from the laser to the analog-to-digital converter (Fig. 4), deriving an equivalent digital model, relating the digitized samples to the transmitted bits of the probe signal, and incorporating the effects of the receiver response and the laser pulse shape, on the digital reconstruction of the fiber backscatter response. We start by defining normalized versions o f f ( t) and r (t): The backscatter response f( t ) is normalized to an initial unity value, by dividing by its initial value:

coded according to a digital sequence pk which takes the values 0, 1 . Let Eop0(t) be the response of the laser driver to a single digital pulse, pk = 8 k , with Eo the energy of the laser pulse response, i.e., the laser pulse response bo is normalized to unit area:

1
I--

Do(t) dt = 1.

(30)

-W

Assuming that the laser and its driver represent a digitalto-analog linear time invariant system, the output of the laser when driven by the digital sequence p n is given by a superposition of shifted single pulse responses:
L- 1

p(t) =
=

n=O
L- 1 n=O

pnEoBo(t - n ~ )
pnS(t - n T ) @

&Bo

(t).

(31)

(24)

Substituting (24), (27), (3 1) in (4b) yields where the normalizing quantity (initial backscatter) is given by
L- 1

~ ( t= ~ ( 0T )( o ) E O ) ~

n=O

C PnS(1

n ~ )

For an ideal fiber with uniform losses, the normalized backscatter waveform is given by

Let us introduce the O

~ response: R

(26) The receiver impulse response and the laser pulse shape are normalized to unit area:
f(t) =

w ( t ) = P ( t ) (0

Po ( t )

(33)

combining the effects of the laser and the receiver, and he smoothedjfiber response

r ?(t) = where

v)

0)

h&)
(27 )

= w(t)

8.m

(34)

R(f) =

which represents the backscatter response filtered through the receiver and smoothed by the single pulse laser response. The output voltage may then be written as:
L- 1

r(t)e-j2wfidt

-m

n=O

is the transimpedance frequency response, and

L-1

R(0) =

ca

= ~ ( 0T )( o ) E O ~

r ( t ) dt

n=O

Pnhf(f - n ~ ) *

--OD

(29)

(35)

is the dc transimpedance. We now assume a laser probe signal p ( t ) which is

Notice that the absolute level of the detected voltage is given by the product of the pulse energy Eo, the initial backscatter f ( 0 ) (equation (25)), the coupling and re-

30

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 7. NO. I . JANUARY 1989

sponsivity factor 7, and the receiver dc transimpedance R ( 0 ) (equation (29)). It is apparent that the output voltage is obtained as a superposition of OTDR responses, each of which is generated with the appropriate delay n T corresponding to one element p n of the transmitted sequence. An example of a received voltage waveform is shown in Fig. 5 . An analog-to-digital converter digitizes the voltage at fixed T intervals. Neglecting quantization noise, the measured digital samples are given by
L-1

10

20
time

30

40

Fig. 5 . Received voltage waveform. Each bit of the probe signal generates a backscattering signal. It is their superposition that is detected and processed.

xk

u(kT) = R(0)qf(0)Eo

n=O

p n h f ( k T- n T ) . on the performance of actual systems. The OTDR resolution is determined by the width of the OTDR response (equation (33)) which becomes the autocorrelation of the laser pulse shape for a matched filter receiver, under the substitution (equation (42)):
w ( t ) = Po( - t )

(36) Let us denote the samples of the OTDR response by hk:


hk = h f ( k T ) . (37) The measured digital samples (equation (36)) may be rewritten as:
L- 1

8 PO(t) = PoO) * P o ( t ) .

(43)

xk = 20 where 26
=

n =O

c Pkhk-n

(38)

N O ) 7f(O)Eo.

(39)

Defining an equivalent digital input uk = 2ppk (40) the whole process of injection of pulses into the fiber, detection and amplification of the backscattered, and its digitization is reducible to an equivalent discrete-time convolution between the equivalent input sequence uk (equation (40)) and the samples hk of the OTDR response (Fig. 4(b)):
L- I

Thus, for a matched receiver OTDR, the inverse of the t) autocorrelation width of the transmitted laser pulse may be quoted as the system resolution. For rectangular pulses of duration Tp the half width full maximum of the triangular autocorrelation is Tp,thus the resolution is given by T i l . On the other hand, the gain factor (equation (39)) for the backscattering is proportional to the laser pulse energy, which in turn is proportional to the pulse duration, when peak-power limited. It follows that a direct trade-off exists between sensitivity and resolution, as discussed in Section 11. The sensitivity-resolution product may, however, be improved by encoding the transmitted signal and designing appropriate processing schemes as explained below.

eo(

xk =

n=O

c Ukhk-,, =

uk

8 hk.

(41)

This formulation incorporates the analog electronic and optical effects into the various normalizations, reducing the problem to a digital signal processing one. The signal processing challenge is to design the probing sequences uk and processing algorithms making possible the extraction of the OTDR response hk from the received voltage samples xk by digital signal processing. If either the laser pulse or the receiver are bandlimited to 1 / 2 T , the analog waveform h , ( t ) may be retrieved from the samples hk. However, h f ( t ) differs from the actual backscattered response f ( t ) by the analog windowing effect of the laser and receiver responses. For a given laser pulse shape p o ( t ) and in the presence of white noise, it follows from matched filter theory that the best SNR is obtained when the receiver impulse response is matched to the transmitted laser pulse, i.e.

(42) Although the matched receiver may be hard to achieve in practice, its value lies in providing a useful upper bound
i ( t ) = Po(-t).

VI. UNIPOLAR SIGNAL PROCESSING Suppose the impulse response hk of a linear system is to be identified, but the physical constraints preclude the direct injection of bipolar codes into the system. This is the case with a direct-detection OTDR wherein only unipolar optical intensity signals are available for detection. In this section we present a novel technique adapting bipolar complementary code system identification techniques to the processing of return signals from intensity modulated OTDR probe signals. The solution is to transmit the bipolar codes on a bias which is equal to half of the available peak power. The fiber is probed by successive groups of four probe shots, this sequence being comprised of each member of an L-bit Golay code pair and its ones complement. By subtracting the measured backscattering signals in pairs, the component of the signals generated by the bias is cancelled, whereas the component generated by the codes is reinforced. By correlating these differences with their respective codes and summing the correlation results, an estimate of the fiber response is obtained. Given a bipolar code ck = + p , where p is a constant,

NAZARATHY et al.: LONG RANGE COMPLEMENTARY CORRELATION OTDR

31

one may associate with it a unipolar code

Notice that in the OTDR problem the constant level 0 is given by (40). The unipolar code is simply related to the bipolar code by a vertical shift: (44b) Let i i k be the unipolar 1s complement of uk, obtained by swapping the high and low values:
=

uk =

0 + ck.

uk13 k hk Yk

(b) Fig. 6. Bipolar processing with unipolar signals. (a) A bipolar signal is mathematically synthesized at the input as the difference of two unipolar signals. (b) Equivalently, the two unipolar signals are individually injected at the input and the outputs are subtracted.

[0,
0,

uk =

-0
0.
(444

uk =

Then the 1s complement may be expressed in terms of the associated bipolar code as (444 Subtracting (44b) and (44d) it becomes apparent that a bipolar code is expressible as a difference of unipolar codes: (45) Applying a linearity argument it follows that if the system is sequentially probed, first with the unipolar code uk:
Xi1)

uk =

0 - ck.

ck = u k

- iik.

The subtraction of partial outputs (equations (5 1)) cancels the outputs due to the bias term bk but reenforces the outputs due to +ck, resulting in (50). Combining this unipolar signal processing technique with bipolar complementary code system identification, and applying the resulting scheme to optical time domain reflectometry is our novel contribution to overcoming the problem posed by the nonnegativity of light intensity of correlation OTDRs. The detailed algorithm is described in the next section. VII. COMPLEMENTARY CORRELATION ALGORITHM Coaching the system identification scheme in OTDR terms, the algorithm is as follows (Fig. 7): 1) Four probe signals are injected into the fiber and the backscatter returns detected: a) a unipolar version of the first code of a Golay pair, b) its 1s complement, c) a unipolar version of the second code of a Golay pair, d) its 1s complement.
2 ) The returns from the first two signals are subtracted from each other and correlated with a bipolar version of the first Golay code. 3) The returns from the last two signals are subtracted from each other and correlated with a bipolar version of the second Golay code. 4) The correlation outputs are added up to yield an estimate Y k of the fiber response hk.

= hk

8 Uk

(46) (47 1

then with the complementary unijdar code i i k :


.Ti2)

= hk (8 u k

and the results subtracted:

the procedure amounts to having tested the system with the difference of unipolar codes: (49) which amounts to testing the signal with a double amplitude bipolar code: Xk = 2hk (8 ck. ( 50a )
xk

= hk (8 [uk

- iik]

This procedure is shown in a block diagram form in Fig.

6.
An alternative interpretation may be obtained by viewing the unipolar codes (equations (44b) and (44d)) as bipolar riding on a bias sequence bk, given by:
b, 0,

O i k c L
otherwise.

Injecting uk and i i k into the system under test amounts to injecting the bipolar codes f ck, respectively, superposed on the box-car sequence:
Xr)

The signals generated in each of the stages of the algorithm are outlined below: 1) The injected signals: a) U - = @ ( I Ak) : b) U = p(1 - Ak) : c) - = p(1 Bk) d) U = p(1 - Bk). :

hk (8

Ck

-k hk (8 b k

(514

2) The subtracted returns due to code A and its complement:

32

JOURNAL OF LIGHTWAVE TECHNOLOGY. VOL. 7 , NO. I . JANUARY 1989

Fig. 7. Complementary code OTDR algorithm. The fiber is sequentially probed with two unipolar versions of codes A , B and their one's complements. Results are subtracted in pairs, correlated with the respective codes, and summed.

The correlation with code A : y$


=X t

* Ak.
-

3) The subtracted returns due to code B and its complement:


Xf =

VIII. SIGNAL-TO-NOISE ANALYSIS In the time it takes a conventional single-pulse OTDR to make 4 measurements, the coded probing scheme effectively makes 2L measurements, with every transmitted bit contributing to the result, amounting to simultaneously measuring many backscattering curves, each generated by an individual bit. As Fig. 5 indicates, the raw signal that is detected is the superposition of these backscattering curves. While the raw signal is unrecognizable as the fiber response in this form, the signal processing effectively rearranges the superposed signals to recover the fiber impulse response with the resolution of a single bit. At the same time, the extra energy that is injected into the fiber and subsequently detected results in a signal-to-noise improvement of the order of By the end of 4N shots, each of the two complementary codes and their complements have been averaged N times, and the data is transferred to perform two correlations with the respective complementary codes. In the absence of noise the final signal estimate is:

A.

hk
B

8 U:
=

hk 8 U f .

Yk

= 4wOhk.

(53)

The correlation with code B: yk xk


B

* Bk.
+Y .:

4 ) The added correlation outputs:


Yk =

Following the reasoning of the last section, the partial output differences are indeed the fiber responses to bipolar Golay test signals:
X$ =

20hk 8 Ak 20hk

(51a) (51b)

Xf =

8 Bk.

The averaging, correlation, and subtraction operations all involve signed additions of digital samples. Thus the noise fluctuations add on an rms basis, i.e., grow as the square root of the number of signed additions. In the course of N averages the rms noise fluctuations in each sample grow as f i ,the subtraction of samples contributes a factor h noise buildup factor, and finally correlation contributes a & factor and the final addition contributes a h factor. Altogether, the noise builds up by a factor of G L . Let ou be the rms noise of the voltage samples entering the analog-to-digital converter. The SNR improvement in the kth sample after 4N cycles, after the signal has been processed as described above is given by SNR;dd = Yk -

The correlator outputs are


Y f = 2Ohk 8 Ak

* Ak * Bk

(51c) (51d)

(Tu

&

h & E0 k
uu

(54)

y: = 20hk 8 Bk and the sum of correlations is

where (53) was used in the second equality. This is to be compared with a conventional OTDR scheme which makes use of single pulse injection with the same peak power as the coded scheme described above, modeled as: uk = 206k
=

R(o)Vf(O)E$k

(55)

This technique may be combined with signal averaging to yield maximum improvement in the SNR in a given measurement time. Since averaging is a linear operation which commutes with the linear operations of summing and subtraction of signals and with correlation, it follows that averaging may be performed either before or after the correlation, without affecting the end result. Since the correlation operation may be time consuming it is most advantageous to minimize the number of performed correlations by first averaging many identical repetitive shots and finally applying a single correlation (for each complementary code) to the averaged result.

where (39) was used. After 4N averages, the signal builds up to a value:
Yk =

4Nhk 8

Uk

= 8NOhk

(56)

whereas the noise becomes C N a v , thus the SNR in the kth sample for a single pulse OTDR is given by:
S N R ~ I S ~ ~

J4n

Yk

=4 J - A
Iu s

Oh

(57)

0 ,

The overall SNR improvement with respect to a conventional averaging scheme which processes the same number of shots in the same measurement time, is given by

NAZARATHY et al.: LONG RANGE COMPLEMENTARY CORRELATION OTDR

33

OTDR is obtained by using a code length of 4 (Lo,, = 2). From that point on, each doubling of the code length improves the sensitivity by 1.5 optical decibels with respect to the single pulse case. Notice that doubling the laser pulse energy Eo improves the sensitivity by a factor of 3 IX. SIGNAL-TO-NOISE IMPROVEMENT dB. The range z in kilometers at which a given sensitivity The last section examined the relative SNR perfor- is attained increases by 1.5 dB/2a [dB/km], for each mance of single pulse versus complementary code OTDR. doubling of either averaging time or code length. AlterL means that avThis section investigates the effect of correlation on the natively stated, the dependence on & key performance parameters of an OTDR, deriving de- eraging time and code length may be directly traded off, tailed expressions for the absolute sensitivities of the two i.e., doubling code length reduces by :: factor of 2 the measurement time required to achieve a certain accuracy. schemes. The root-mean-square noise voltage q,at the input of This advantage is shown in Fig. 8. By solving (60a) for the analog-to-digital converter may be expressed in terms the number of probe shots and assuming a l-kHz repetition rate and 1-ps bits, the time required to achieve an of the noise equivalent power P N E as: amplitude sensitivity of 0.1 dB (SNR = 17 dB ) is plotted cu = R(o)?PNE* (59) as a function of 2-way range. The curves are plotted for Substituting the last equation as well as (39) in (54), as- a conventional OTDR, and for a CCOTDR using code suming an ideal uniform fiber, and switching to a variable lengths of 16, 64,and 256. z representing the one-way distance into the fiber, yields: The performance advantage of the CCOTDR can be seen in two ways. By comparing the conventional result with the result of a 256-bit code measurement after 60 s of measurement time, the CCOTDR is seen to measure the same sensitivity 9 dB farther into the fiber. In practice or in optical decibels (10 log,,, P ) (two-way): this might correspond to a range improvement of 10-15 S N R , (~ ) ~ Z km, depending on fiber loss. Alternatively, the measurement range and amplitude sensitivity obtained with a con=f(0)idB1 + - pNE[al ventional OTDR in one minute of measurement time (0.1dB sensitivity at 30.5 dB) is obtained with the CCOTDR + 1.5 dB(N,, + Lmt) - 2z[km]a[dB/km] using 256-bit probe codes in less than I s. The performance advantage of a CCOTDR over a conwhere No,, and Lo,, are Nand L expressed in octaves, e.g., ventional single-pulse OTDR using the same laser, receiver, and other components can be summarized as folNo,, = log, N . Similarly, lows. When coded probe signals consisting of L bits are used to probe a fiber, a two-way dynamic range increase NE of is achieved relative to a conventional singlepulse measurement in a given averaging time, while or in decibels maintaining the response resolution corresponding to the SNRPU~S~ (4 width of a single bit. Alternatively, the same result obtained with a single-pulse measurement can be obtained a =f(o)[dB1 + - PNE[dBl factor of L / 4 times faster using complementary codes. It 1.5 dB(Nmt + 2) - 2 ~ [ k m ] ~ ~ [ d B / k m ] . follows that a conventional OTDR measurement which typically lasts for one minute is completed in less than one ( 60b ) second when a code length L 1 256 is used. It is in this These equations generalize the conventional OTDR sense that CCOTDR systems are designated as real makers formula ( 5 ) to describe the performance of the time. complementary correlation OTDR (CCOTDR) . This discussion has not taken into account signal proIt has been shown that the correlation technique allows cessing overheads associated with performing digital corL more effective averages in the same measurement time. relation. Generally, an SNR improvement is associated To make a fair comparison, however, it should be noted with performing correlations, however the time spent in that the conventional probe is a single pulse at the full doing digital correlations extracts an SNR penalty, since peak power of the laser, whereas for the coded probe each the same time could be spent doing averaging in a conbit has an amplitude of only half of the peak power due ventional system (unless digital pipe-lining is applied in to the bias that it must ride on. In terms of the OTDR- which case the full measurement time may be spent avmakers formula, this means that P,,,, is 3 dB larger for a eraging). The overall performance is finally determined conventional measurement than for a coded one that uses by the balance between these two effects. In the current the signal processing scheme described previously. As a scheme, the use of only a pair of Golay sequences, rather result, performance equivalent to that of a conventional than a large number of complementary sequences [15],

the ratio of (54) and (57):

34

JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL 7. NO. 1 , JANUARY 1989

YH-1
GENERATOR

FIBER UNDER TEST


~

P
CLOCK

L3-5
COUPLER

E-WRY

RRNGE ( d B )

Fig. 9. Block diagram of the CCOTDR system.

Fig. 8. Time to measure 0.1-dB faults versus two-way range for a conventional OTDR and a CCOTDR using code lengths 16, 64, and 2 56.

results in a considerable reduction in correlation time overhead by taking advantage of the commutability of the operations of averaging and correlation, and performing only a pair of correlations every so many averages. In fact, it is possible to completely eliminate this processing time penalty by using pipelining in the digital processing block, an option that would be rendered difficult or impossible if a large set of autocorrelations had to be performed. The following section describes how these performance improvements were demonstrated in practice.

speed processing capability allowed the reconstructed data to be displayed on a CRT with updates of the averaged data approximately every 250 ms. A variety of fiber configurations were used to test the performance of the CCOTDR. The fibers used in these experiments had attenuation coefficients ranging between 0.34 and 0.4 dB/km. Elastomeric splices were used to make most of the connections, although measurements made on connectorized fiber yielded similar results.

X. EXPERIMENT
A . Setup A block diagram of the experimental arrangement used to make the CCOTDR measurements [24] is shown in Fig. 9. A code generator was used to drive a commercially available InGaAsP laser diode ( A = 1.3 pm) with the appropriate Golay codes and their ones complements. The shot-to-shot repetition rate of the code bursts was approximately 700 Hz. A peak power of 6 mW was coupled from the laser into its single-mode fiber pigtail. The laser power was coupled into the fiber under test using a fiber directional coupler whose second output was index matched to suppress reflection. The peak power coupled into the fibkr under test was approximately 2 mW. Approximately half of the return signal from the fiber under test was coupled to the receiver via the directional coupler. The receiver consisted of a pigtailed InGaAs p-i-n photodiode followed by a trans-impedance amplifier. The bandwidth of the receiver was approximately 3.5 MHz, and its noise equivalent power (NEP) was 400 pW (-64 dBm). The amplified signal was then digitally sampled, averaged, and processed to reconstruct the fiber backscattering impulse response. In the first experiments, the averaging, correlation, and display were performed using a desktop computer (HP9836). For the later experiments, including many of those described in this paper, the functions of the clock, code generation, averaging, and correlation were performed using a specialized 32-bit 10-million-operations-per-second signal processor. This high-

B. Results The results presented here represent the key performance advantages of this complementary correlation technique: improved dynamic range, and greatly reduced measurement time, signal-to-noise improvement without compromising resolution, and strong sidelobe suppression. Fig. 10 shows a measurement of an extremely large reflection made using 32-bit codes made up of 4-ps bits. This result, which is plotted on a linear scale, shows autocorrelation sidelobes which occupy a fraction of one division to the right of the peak. However, since the height of the peak is 1187 of these divisions, the worst sidelobe deviation (which could also be interpreted as receiver undershoot) is 38 dB down, with the rest of the structure more than 40 optical decibels below the peak (corresponding to 80 electrical decibels). It should be noted that this is a typical rather than a best result-in some cases sidelobe suppression approaching 45 dB has been observed. This is the best sidelobe suppression reported to date for a correlation OTDR and is 20-30 dB better than what is possible using Barker or finite pseudorandom codes, even under ideal conditions. Fig. 11 shows two measurements of a 5-km span of fiber centered at 25 km after 5 s of averaging time. In both cases, the bit duration was 125 ns. In Fig. ll(a), a 4-bit code was used to represent the result that would be obtained using a conventional OTDR. Fig. ll(b) shows the result of using 256-bit codes in the same 5-s measurement time. In this case, the noise is greatly reduced and the backscattered signal is visible. This result is obtained without sacrificing resolution, as evidenced by the fact that in both cases the 3-dB width of the reflection peak is only 16 m.

NAZARATHY er al.: LONG RANGE COMPLEMENTARY CORRELATION OTDR


- 6 5 . E dBm

35

- 4-- t - -1-

- 1 - -1-

- 4 - - c - 4- - t - -

MICROSECONDS

160

Fie. 10. Measured reconstruction of a laree reflection. Most sidelobes are more than 40 dB down fmm the peak.

--

STRRT POINT: 35.68 km SPAN: 1.E km VERTICRL SCRLE: 2 . 5 d B P e r D i v ( o n e way)

Fig. 12. A 26-meter resolution at a distance of 35.4 km. Measurement time: 20 s.

- 5 4 . 0 dBm

-94.8 dBrn STRRT POINT: 2 2 . 5 0 km SPRN: 5 . 0 km VERTICRL SCALE: 2 . 5 dB Per D i v (one uay) CODE LENGTH: 4

-54.0

dBrn

-94.0

dBm

Fig. 11. Two 5-s measurements using 125-ns bits. Upper curve represents a conventional measurement. Lower curve results from using 256-bit codes.

An example of high resolution at a long range is shown in Fig. 12. The reflection feature is located at a distance of 35.4 km. It contains not one, but two peaks that are separated by only 26 m and clearly resolved. These peaks are the result of reflections from both ends of a 26-m segment of fiber which is spliced into the cable at that distant point. Whereas a conventional OTDR using the same receiver, laser pulsewidth and peak power might take well over 20 min to obtain such a result, this measurement, which used 256-bit codes, took only 20 s.

Fig. 13 shows two measurements of a 10-km span of fiber centered at 30 km after 10 s of averaging time. In both cases, the bit duration was 250 ns. In the measurement shown in Fig. 13(a), wherein a single-bit code was used, a reflection is visible at about 27 km and the backscattering signal drops into the noise. When a code length of 128 was used (Fig. 13(b)), however, the backscattering signal is free of noise and the end of the fiber is clearly 10.5 dB in the seen. The noise level was reduced by same measurement time, in good agreement with theory ( 7 octaves X 1.5 dB /octave). Fig. 14 shows two measurements of the same 20-km segment of the backscattering response of a fiber recirculating delay line [2]5. Both measurements exhibit roughly the same signal-to-noise ratio. However, whereas the measurement using 1-bit codes required 256 s of averaging to obtain this result, the same result was obtained using 32-bit codes in only 8 s, or 32 times faster. Finally, Fig. 15 shows three measurements of a 20-km span of fiber between 20 and 40 km from the input. In each case, 500-ns bits were used. The measurements shown in Fig. 15(a) and (b) were made using 4-bit codes to represent the performance of a conventional singlepulse OTDR. After 15 s of averaging (Fig. 15(a)), a single reflection is visible at 25 km, and the backscattering signal descends into the noise. After 16 min of averaging (Fig. 15(b)), the backscattering signal is largely free of noise, a second reflection is visible at 31 km, and the end of the fiber at 35.5 km is clearly seen. However, by using codes of length 256 (Fig. 15(c)), the same result was obtained after averaging only 15 s. Relative to the 15-s measurement made with 4-bit codes (Fig. 15(a)), the two-way dynamic range is improved by 9 dB. These experiments verify each of the major performance advantages predicted by the theory of the complementary correlation OTDR: low autocorrelation sidelobes, improved dynamic range in a fixed measurement time, and a significantly reduced measurement time at a given range. Furthermore, these improvements are accomplished without any sacrifice in response resolution. Recently, a new result was obtained using a production prototype of a commercial product under development [26] (Fig. 16). A 62-km length of fiber was measured

36

JOURNAL OF LIGHTWAVE TECHNOLOGY. VOL. 7. NO. I. JANUARY 1989

L=l

91

L=128

1
SPRN: 20 km STRRT POINT: 20.08 km VERTICRL SCRLE: 5 . 0 d B P e r D i v ( o n e way)

..

. .

2 LmIDlV

Fig. 13. Two 10-s measurements of a 10-km span centered at 3 0 km, using 250-11s bits. (a) Single-bit measurement. (b) Measurement using code length 128.

MEASUREMENT TIME: 256 SECONDS


STRRT POINT: 20.88 km VERTICRL SCRLE: 5 . 8 d B P e r SPRN:

114.0 dBm 20 km

Div

(one u a y )

-54.0 dBrn

-i--r --

r--I-- - -I - _ _I _,5iECO& _ _ _ _ _ 1- - 7 - - r
I

L = 256

- -1- _ L _
Fig. 14. Two measurements with equivalent SNR. 500-ns bits were used in both cases. The 32-bit measurement was made 32 times faster.
START POINT:

-1-

_ I - _ I _ - 1- - L _

-114.0 dBm 28.00 k m


SPRN:

20 km

VFRTTCRL SCRLE: S . E dB P e r Div ( o n e w a v l

using 32-bit codes with 4-ps bits. The one-way dynamic range of this system was determined to be 30.5 dB, as measured from the initial backscattering level to the level of the rms noise. Previously, the best performance reported for a practical system [16] was observed using a coherent-detection OTDR with 30-dB one-way dynamic range after lo6 averages, with a response resolution of 500 m (using 5-ps pulses) and an amplitude sensitivity of 0.2 dB. The current direct-detection complementary correlation result was obtained after only - 8 X lo5 averages, with a response resolution of 400 m (using 4-ps bits), and an amplitude sensitivity of k0.05 dB. Thus, the sensitivity advantages of coherent detection were exceeded while eliminating drawbacks and stringent requirements for the coherent schemes such as laser source spectral punty, the need for a frequency-shifted local oscillator and polarization noise.

Fig. 15. Three measurements in the range 20-40 km. The upper two measurements were made with the equivalent of a conventional OTDR. The lower measurement used L = 256 codes.

'1

'

D l ~ t . : 81.675 km

c9 .......i...

z
->

: .. . . . . ............................................ . . .. 1 ; :. . . . 1 : . . . . .........................................................
.......................... . . . . .
i . .
: ""I"

i.. ............. :' . . .


..j

e =
U

'

..... j........... . . . i ..... : . i


:

I
I

: : . . . .. . . . . . . . ... . . . .. .

:.
.

Fig. 16. Measurement of 62 km of fiber, using 32-bit codes with 4-ps bits. Dynamic range exceeds 30.5 dB.

XI. CONCLUSION This paper presented the theoretical foundations and experimental results for a new correlation OTDR technique which yields improved dynamic range, and greatly reduced measurement time, without compromising resolution. The introduction of a novel signal processing algorithm which handles unipolar aperiodic codes, was essential to the application of complementary correlation methods to direct-detection OTDR's. Unlike many highperformance OTDR results reported in the literature, the

technique described here can be used in a practical, rugged, field portable instrument. This is evidenced by the recent introduction of a commercial product which makes full use of the new technique [26]. Furthermore, complementary correlation coding is compatible with future advances in OTDR laser sources and receivers, and may be used to augment future coherent OTDR designs to further improve their performance. ACKNOWLEDGMENT The authors would like to acknowledge the efforts of others who have contributed to this work. They are grateful to M. Fleischer-Reumann, J. Vobis, W. Pless, S .

NAZARATHY

er

U/.:

LONG RANGE COMPLEMENTARY CORRELATION 01TDR

37

Gross, J. Rivoir, J. Beck, and P. Zorabedian for their efforts.


REFERENCES
[ I ] P. Healey, Review of long-wavelength single-mode optical fiber reflectometry techniques, J . Lightwave Technol. , vol. LT-3, pp. 876886, 1985. 121 P. Healey, Instrumentation principles for optical time domain reflectometry. J . Phys. E: Scientific Instrum., vol. 19, pp. 334-341, 1986. 131 Bruce L. Danielson, Optical time-domain reflectometer specifications and performance testing, Appl. Opt., vol. 24, pp. 2313-2322, 1985. [4] E. Brinkmeyer, Analysis of the backscattering method for single mode optical fibers, J . Opt. Soc. Amer., vol. 70, p . 1010, 1980. [5] M. Nakazawa, M. Tokuda, K. Washio, and Y. Morishige, Marked extension of diagnosis length in optical time domain reflectometry using 1.32-pm YAG laser, Electron. Lett., vol. 17, p. 783, 1981. [6] K. Noguchi, Y . Murakami, K. Yamashita, and F . Ashiya, 52-kmlong single-mode optical fiber fault location using the stimulated Raman scattering effect, Electron. Left., vol. 18, p. 41, 1982. [7] Martin P. Gold, Design of long-range single-mode OTDR, J . Lightwave Technol., vol. LT-3, pp. 39-46, 1985. [8] F. E. Nathanson, Radar Design Principles. New York: McGrawHill, 1969. [9] K. Okada, K. Hashimoto, T. Shibata, and Y. Nagaky, Optical cable fault location using correlation technique, Electron. Lett., vol. 16, p. 629, 1980. [IO] M. Zaboli and P. Bassi, High spatial resolution OTDR attenuation measurement by a correlation technique, Appl. Opt., vol. 22, p. 3680, 1983. [ l l ] A. Sudbo, An optical time domain reflectometer with low-power diode lasers, J . Lighrwave Technol., vol. LT-I, pp. 616-630, 1987. [I21 J. J . Bernard, J. Ducate, Y. Gausson, J . Guillon, and G . Le Blevennec, Field portable reflectometer for single-mode fiber cables, in Proc. 10th Eur. Conf. Opt. Commun. (Stuttgart, W. Germany), 1984. 1131 J. I . Bernard, I . Ducate, Y. Gausson, J . Guillon, and G. Le Blevennec, 1.3 pm portable reflectometer for field test of single-mode fiber cables, in Tech. Dig. Symp. Optical Fiber Measurements. Washington, DC: NBS, 1984, pp. 95-98. [14] P. Healey, Pulse compression coding in optical time domain reflectometry, in Proc. 7rh Eur. Con& Opt. Commun. (Copenhagen, Denmark), 1981, p. 5.2-1. [I51 P. Healey, Optical orthogonal pulse compression codes by hopping, Electron. Lett., vol. 17, pp. 970-971, 1981. [I61 J . P. King, D. F. Smith, K. Richards, P. Timson, R. E. Epworth, and S . Wright, Development of a coherent OTDR instrument, J . Lightwave Technol., vol. LT-5, p. 616, 1987. 1171 M . J . E. Golay, Complementary series, IRE Trans. Info. T h r o n , vol. IT-7, p. 82. 1961. [IS] M. J . E. Golay, Multislit spectroscopy, J . Opt. Soc. Amer., vol. 39, pp. 437-444, 1949. [19] G. F. Franklin and D. Powell, Digital Control ofDynamic Systems. Reading, MA: Addison-Wesley, 1980. [20] J . Borish and J . B . Angell, An efficient algorithm for measuring the impulse response using pseudorandom noise, J . Audio Eng. Soc., vol. 31, pp. 478-488, 1983. [2 I ] B . B. Lee and E. S . Furgason, Pseudorandom codes for multi-mode operation of phased-arrays, in Proc. 1980 Ultrasonics S y m p . , 1980, pp. 941-944. [22] B . B. Lee and E. S . Furgason, Golay codes for simultaneous multimode operation in phased arrays, in Proc. 1982 Ultrasonics Symp., 1982, pp. 941-944. (231 Y . Takeuchi, An investigation of spread-energy method for medical ultrasound systems, Ultrasonics, vol. 17. pp. 175-181, 1979. [24] S. A. Newton, Moshe Nazarathy, R. P. Giffard, D. S . Moberly, F . Sischka, S . Foster, S . Gross, and P. Zorabedian, Spread spectrum optical time-domain reflectometer, in CLEO 87 Tech. Dig. (Baltimore, MD), 1987, p. 138. [25] S. A. Newton, M. Nazarathy, and W. R. Trutna, Jr., Measured backscattering signature of a fiber recirculating delay line, Appl. O p t . , vol. 25, p. 1879, 1986. [26] Hewlett Packard 8145A OTDR.

Moshe Nazarathy received the B.Sc and D Sc. degrees in electrical engineenng from the Technion, Israel Institute of Technology From 1982 to 1984 he held a postdoctoral appointment with the Information Systems Laboratory at Stanford University, Stanford, CA His thesis and postdoctoral work involved operator representations of slowly varying envelope optics and their applications to Founer optics, holography, optical resonator theory, phase conjugation, dispersion, and diffraction in anisotropic media He joined the Instruments and Photonics Laboratory of Hewlett Packard CO , Palo Alto, CA, in October 1984, where he has worked on fiber optical signal processing circuits and test instrumentation, fiber reflectometry, solid state lasers and ultrafast electrooptic modulation and sampling. Dr Nazarathy is a member of the Optical Society of Amenca.

Steven A. Newton (S81-M83) was born in Teaneck, NJ, in 1954. He received the B.S. degree in physics, summa cum laude, from the University of Massachusetts, Amherst, MA, in 1976. He received the M.S. and Ph.D. degrees in applied physics from Stanford University, Palo Alto, CA, in 1978 and 1984, respectively. His work as a Research Assistant at Stanfords Ginzton Laboratory involved fiber-optic rotation sensors, single-mode fiber components, and fiberoptic signal processing devices. From 1978 through 1982, he was a part-time-Member of the-Technical Staff at Hewlett-Packard Laboratories, Palo Alto, CA, where he was engaged in research involving optical design, metal vapor lasers, and optical data storage. Since 1983, when he became a full-time Member of the Technical Staff at Hewlett-Packard Laboratories, his research activities have involved photonic measurement systems, optical fiber components and circuits, and integrated optics. Since 1986, he has been in his present position as project manager of the fiber optics group in the Photonics Technology Department of HPs Instruments and Photonics Laboratory. Dr. Newton is a member of the Optical Society of America.

*
Robin P. Giffard (M75) is a native of Guernsey, U.K. He received the degrees of B.A. and D.Phil in physics from Oxford University. After working on low noise amplification using Josephson Effect devices at Stanford University, he moved to Hewlett Packard in 1980. Since then he has worked on precision instrumentation and atomic frequency standards.

David S. Moberly received the B.S. degree from the Massachusetts Institute of Technology in 1975. Upon graduation he worked for Digital Equipment Corporation designing PDP- 11 minicomputers. He joined the Hewlett-Packard Company in December 1979 where he has worked on computer system design, graphics, and digital signal processing. Mr. Moberly is a member of the Association for Computing Machinery.

38

JOURNAL OF LIGHTWAVE TECHNOLOGY. VOL 7. NO I . JANUARY l Y 8 Y

Franz Sischka was born in Stuttgart, West Germany, in 1954. He received the Diplom degree in 1979 and the Dr.-Ing. degree in 1984 from the Universitat Stuttgart. From 1979 to 1984 he worked on the modeling of bipolar transistors with SPICE and the design of low power operational amplifiers. He joined Hewlett-Packard GmbH in Boblingen, West Germany, in 1984 and is presently working on instruments for fiber optics.

degrees from Stanford University in 1974 and 1978. respectively. From 1978 to 1980 he was a Guest Researcher at the Max-Planck Institute for Quantum Optlcs in Garching, W . Germany. In 1980. he joined Hewlett-Packard Co.. where he is currently a Project Manager at Hewlett-Packard Laboratories. He has been active in the research of solid state lasers, tunable semiconductor lasers, nonlinear optics, and photolithography.

*
William R. Trutna, Jr. (S71-M77) was born in Pasadena, TX, on September 29, 1951. He received the B.S. degree in electrical engineering from the University of Texas at Austin in 1973, and the M.S. and Ph.D.

*
S. Foster, photograph and biography not available at time of publication.

You might also like