An Experiment In Sequence Time Domain Reflectometry For Fiber Optic Diagnostics

Sequence TDR is an alternative approach to time domain reflectometry that modulates a carrier with a pseudorandom code rather than an impulse or a step or any simple recurring waveform. The benefit is that returning reflections encoded with such a pattern auto-correlate very well when the codes line up and cross-correlate negligibly with other portions of the code sequence. Correlation gives rise to higher signal-to-noise ratio than other forms of reflectometry because high duty cycle contributes more energy to the result. This phenomenon relates to correlation gain in code division multiple access (CDMA) communication systems and is known to RADAR practitioners as a pulse compression scheme.

Two professors capitalized on their efforts at Utah State University and the University of Utah to found a company called LiveWire Test Labs¹ that develops and markets Sequence TDR and Spread Spectrum TDR systems for detection of electrical wiring faults. Such systems use radio frequency (RF) signals modulated by code sequences to perform reflectometry. Professor Cynthia Furse said they considered STDR for fiber optic systems but had not yet pursued development for this market.²

I predicted that STDR would a good candidate for fiber optic applications and also be a good choice as a built-in diagnostic test for fiber optic data link health monitoring in avionic systems. I began work in December, 2010 to see if I could implement such a system and demonstrate the potential benefits. This report details how I applied Sequence Time Domain Reflectometry to fiber optics. My goal is to build reflectometer diagnostics into transceivers so that fiber links can analyze their own faults.

An Approach to STDR for Fiber Optics

Smith, Furse, and Gunther³ launched RF signals encoded by a pseudorandom code along a transmission medium and correlated the returns with increasingly delayed versions of the same code. I designed and built a pair of identical pseudorandom code generators with common clock and reset inputs and devised various means to vary the delay of the second code with respect to the first. I designed this circuitry into an Altera EPM3032/64 complex programmable logic device (CPLD) to achieve much higher clock speed than attainable with logic chip families on hand. I began with a kit provided by Altera 10 years ago and built a test board. This setup allowed me to program the CPLD via an interface cable that connects to a JTAG connector on the board. Altera⁴ provided programming software called Max+Plus II and the interface cable called a ByteBlasterMV contained in the kit.

Figure 1 shows the functional block diagram. The reference code from PN Code Generator 1 drives a fiber optic transmitter at 1310 nm. Radiation passes to one leg of a 3 dB fiber optic coupler and into the common fiber. Half of the radiation is lost as the coupler splits the radiation between the common fiber and an internal node. Reflections from the open fiber end return back along the common fiber to the coupler and split between the transmitter and receiver. Signals from the fiber optic receiver are applied to a passive double balanced mixer that serves as a correlator. PN Code Generator 2 provides successively delayed versions of the same PN code to the other side of the correlator. Each correlation results in a point on a characteristic waveform that depicts the time behavior of the various reflections and thus the contributions versus distance along the fiber. This is exactly analogous to what Reference 3 calls Sequence Time Domain Reflectometry except that 1300 nm radiation serves instead of RF.

1. The functional block diagram illustrates the test setup for the STDR diagnostics.

I subsequently learned that this method is nearly identical to expired Hewlett-Packard patent 5,000,568 for a fiber optic time domain reflectometer. HP used Golay codes instead of simple pseudorandom codes with the additional benefit of suppressing pattern artifacts.

Figure 2 shows the circuitry programmed into the CPLD for an early approach to the problem. Both generators produce the same maximal length pseudorandom code. This example has pseudorandom code generators that are nine registers long and produces identical codes of 511 chips in length (2⁹-1). The code sequences initiate in unison from common start and clock inputs. Both have ‘frame synch’ circuitry to show when the code restarts. Complementary outputs drive the correlator and the fiber transmitter.

2. The circuitry programmed into the CPLD for an early approach includes dual 9- bit PN code generators with common reset and clock inputs.

Circuitry in the upper left corner initiates a sweep when the ‘StartOnRise’ signal causes an impulse to clear the ‘Stop after one code sequence’ Dflipflop and terminates the sweep with an end of sweep signal from ‘FrameSynch1Out’. Copying the elements of this figure into the Altera Max+Plus II software should yield identical operation upon programming into an EPM3064 CPLD.

Circuitry in the lower left corner delays subsequent repeats of the 2nd pseudorandom code sequence with respect to the first in order to correlate with reflections from increasingly greater distance. In this example, the relative delay between successive code repetitions is one full chip. More complex implementations delay successive code repetitions by ½ of a chip or ¼ of a chip. Time resolution ultimately equals the chip period, which derives from the clock frequency, but sampling more often gains data that aids in interpolating to achieve fractional chip resolution. Furse discusses such a method of interpolation.⁵

I achieve half chip delay by advancing the second code in half clock cycle delays. The method is to invert the clock waveform to the second PN code generator on alternate half cycles and halve the count of sweeps so that the full chip delay advances every second sweep while the clock inversion provides half clock cycle delays in between. This works and provides a measurement resolution slightly better than 10 inches at 200 MHz clock speed.

I achieve quarter chip delay and 5-in. resolution by routing two of the four clock phases through an external delay line. This is likely the best that the 200 MHz EPM3064 CPLD can achieve.

Subsequent versions derive the ‘StartOnRise’ signal that initiates each code sequence from the 200 MHz clock by dividing it by 100 or 200. In the divide by 100 case, a code sequence occurs every 50 µs, so the samples in the correlation waveform that represent 5 in. of fiber distance appear 50 µs apart on an extended time scale rather than nanoseconds apart. Sample spacing in time depends on the rate at which code sequences recur. This enables the convenience of relatively low speed analog to digital conversion.

Figure 3 shows correlation waveforms from early measurements. The reference open fiber causes a 4% reflection at 3.25 divisions. An additional fiber length of 42 inches causes a 4% reflection at 4.1 divisions while the reflection at the connection between the two fibers at 3.25 divisions disappears. The additional delay of 0.85 divisions corresponds to 8.5 samples of 4.94 inch resolution.

3. This screenshot shows correlation waveforms on expanded time scale with and without 42-in. fiber extension.

I created an experimental test setup to evaluate the Fiber Optic Sequence TDR diagnostic tool. Results show that FOSTDR is an approach to fiber reflectometry that we can integrate into transceivers. Any drawbacks such as fiber splitter losses also apply to other reflectometry methods. Range of this version is 195 ft with about 18 ft wasted on internal fiber. I adjust maximum range by setting code length.

Noise and Nonlinearity

With little light returning from the fiber end, otherwise unnoticed reflections come into prominence. The two upper traces of Figure 4 show the 4% reflection from the open reference fiber end at about 3.9 divisions and a 4% reflection at 6.5 divisions from an additional 125 in. of fiber connected to the reference fiber end. When the reflection from the second fiber is almost completely suppressed by placing the end into a drop of water, the lower trace appears with increased oscilloscope gain. The delayed (right hand) reflection in the lower trace is the reduced return from the fiber end within the drop of water. The new (left hand) reflection at 3.1 divisions precedes the reference fiber end by 40 in., placing it right at the center of the hybrid splitter. An amplifier increases the PIN-TIA receiver signal before the correlator in this measurement.

4. The upper trace shows 4% reflections from open reference fiber and additional 125 in. of fiber. The lower trace shows how a drop of water suppresses reflections in 2nd fiber and allows a small reflection from the hybrid splitter to appear.

Figure 5 similarly shows the 4% reflection in blue from an open fiber and the much smaller reflections in red from an additional 23.8 inches of fiber with a break at the end. A 12 bit A/D converter acquired these results. Signal averaging reduced noise. The blue 4% peak is the reference end of the fiber from the hybrid splitter and serves to locate the starting point for distance measurements. The first red reflection is due to imperfections within the hybrid coupler, and the number of 1.25 ns samples (5 in.) between it and the 4% peak places it exactly (within ¼ inch) at the center of the hybrid fiber splitter/combiner. The amplitude of the first red peak is 1/10th the magnitude of the open fiber reflection, or perhaps 0.4%. The broken end reflects half of that or perhaps 0.2%. Before breaking the end of the 23.8 fiber, terminating it in a drop of water caused the second reflection on the red curve to go to zero. Breaks result in a range of returns. In this case it is small but visible. The 4% figure derives from the refractive indices of glass and air. The estimates of 0.2% and 0.5% hold only if the response is linear.

5. The large waveform shows a 4% open fiber reflection, while the small waveform shows a 0.4% broken fiber reflection.

Other data suggests the response is not linear. Non-linear behavior is likely due to the double balanced mixer I use as a correlator. I had poor results with a high speed multiplier as a correlator, but that approach will probably be better when it works. The peak from the broken end is about 4.8 samples more distant than the open end, corresponding to its length of 23.8 inches.

The demo hardware resolves reflections to a small fraction of a percent. The lack of reflections from a roughly broken end seemed to present the optimum conditions to try to observe reflections from a fiber with a severe kink or wound tightly around a mandrel. No such reflection is visible under the best conditions.

With orange 62.5/125 µm core/clad multimode fiber in a 1/8 in. hairpin loop or wound tightly around a pencil, there are no visible reflections. Cleaved ends have lower reflection than polished ends, likely because they are not precisely orthogonal to the fiber axis.

Small reflections appear out of the noise when there are no large reflections. Large reflections suppress the smaller reflections. The presence of large signal level reflections from an open fiber connector causes a noise level that masks the presence of much smaller signals from other sources. An amplifier in the signal path improves resolution but also increases non-linearity.

Real System Observation

I expected that good fiber optic receiver design would maximize sensitivity by optimizing photodiodes to absorb every possible photon and thus return small reflections. I tried several SFF and SFP detectors and learned that most have reflections comparable to an open fiber. Only one sample of InGaAs PIN in a ROSA in a Finisar FTRJ1319F1MTL SFF had a very low reflection only twice the height of the return from the hybrid splitter. All others have reflections about equal to the 4% from an open fiber. Most ROSAs apparently lack proper antireflection coatings. Isn’t that interesting?

Noise increases with signal level making it seem like noise in the signal or (2qiB)^1/2 shot noise. It might also be pattern imperfections in the process of correlating ordinary pseudorandom codes that Golay codes are supposed to remedy. To examine this conjecture, I connected the electrical PN code signal that normally drives the optical transmitter directly to the double balanced mixer correlator through various levels of attenuation to observe and compare behaviors.

Figure 6 shows similar results, each again the average of 25 measurements. With all the fiber delay removed, the response occurs much sooner in time. We see:

Waveform shapes have similar identifiable features with some exceptions
Response is not linear – 20dB attenuation reduces amplitude to 1/3
Larger signals have a positive offset
I infer that the noise may be due to the use of a simple code rather than to shot noise

6. These plots show correlation waveforms with direct electrical connection of transmit waveform to the correlator.

I found many papers describing this method and similar methods with other advantages in the literature and patent records. The best of these are referenced at a website titled “The theory and application of fiber optic sensors with spread parameters”⁶. The most important are a paper describing a Hewlett Packard instrument⁷ and recently expired patent 5,000,568.⁸

Conclusion and Recommendation

The results of this effort prove the merits of the Sequence OTDR approach. Future work should expand to a faster programmable logic family to improve resolution further. Experts assure me that some FPGAs achieve 1 GHZ clock rates or better. Such an upgrade will yield 4 in. chips that interpolate to less than 1 inch without the complexities of four clock phases. I would very much like to see someone carry this development forward.

Consider ways to utilize Golay codes. These are out of scope with present hardware. Instead of generating a PN code, one could read repetitive codes out of ROM with an FPGA. I do not know how to generate such a code, and published Golay codes are short.