A high-performance signal-processing system for the HP 8146A optical time-domain reflectometer - includes related article on use of averaging to improve signal-to-noise ratio - Technical

Hewlett-Packard Journal, Feb, 1993 by Josef Beller

Three custom integrated circuits and a powerful 24-bit digital signal processor offload data processing from the instrument's host processor.

The digital signal processing unit of the HP 8146A OTDR acquires and processes data coming from the optical front end after the data has been amplified and converted from analog signals to digital numbers. This data represents the response of an optical fiber under test to a probe from a laser pulse. After further processing and linear-to-logarithmic conversion, the fiber response is transferred to the instrument's host processor and displayed on the OTDR's screen as a function of distance.

The fiber response always depends on two physical effects: Fresnel reflections and Rayleigh scattering. Reflections occur at locations with refractive index discontinuities. Scattering, which is the dominant loss mechanism in singlemode fibers, is generated uniformly along the fiber. One main challenge of an OTDR is the wide range of the input signal level. High power levels occur because of reflections whereas backscattering produces very weak signals which are almost always covered by noise. To detect and identify clearly events that are hidden by noise (to be able to do long-distance measurements), any means of improving signal-to-noise-ratio (SNR) must be applied. Some increase in SNR is achieved by using an avalanche photodiode in the optical receiver instead of a p-i-n diode. Remarkable improvements are possible with digital signal averaging (see "Improving SNR by Averaging," on page 65).

With digital averaging as a standard processing scheme in OTDRs, SNR improvements of up to 30 dB can be obtained during a three-minute measurement. This corresponds to one million averages since the increase of SNR is proportional to the square root of the number of averages. This processing requires a certain amount of hardware because each fiber trace consists of several thousand data samples.

Since the early days of the development and manufacture of OTDRs, engineers have had to struggle with the fundamental OTDR range/resolution trade-off. The high spatial resolution needed in short-haul applications can only be achieved with short laser pulses and a wide receiver bandwidth. This means a low dynamic range. A high dynamic range for long-distance measurements is possible only with long laser pulses and low receiver bandwidth. One of the first questions in the design of an OTDR is where to place the instrument along this range/resolution line and how to balance the properties to fit customer needs in the best manner. It was this question that convinced the HP 8146A development team to build an all-haul instrument that combines both worlds and covers all typical applications.

Digital Averaging and Processing

Digital averaging and processing require sampling and analog-to-digital conversion of the analog signal supplied by the OTDR receiver (see Fig. 1). The analog-to-digital converter (ADC) is one of the key components in an OTDR since its conversion rate and word width influence measurement speed, spatial resolution, and achievable dynamic range. In general, speed goes inversely as resolution (accuracy) at a given cost. In principle, a low-speed ADC with 16-bit resolution would fit into an all-haul OTDR. It provides a high dynamic range for measuring long and 1ossy optical links. On the other end, high spatial resolution can be achieved (if the analog bandwidth is sufficiently high) by applying the interleaving principle (see Fig. 2). However, this resolution is achieved at the cost of measurement time and thereby of noise reduction. To overcome this disadvantage it was decided to use a high-speed, 8-bit flash ADC in the HP 8146A. Its performance is very applicable to short-range applications. To cover long-distance and high-dynamic-range measurements, variable decimation (Fig. 3) and sophisticated stitching (Fig. 4) are implemented to overcome the limited conversion range of the high-speed flash ADC.

With variable decimation two adjacent data samples are summed to one new data point. The geometrical spacing between the resulting data points increases, resulting in a greater measurement span. Decimation or downsampling can be performed with any number of successive samples. An advantage of this method is that it results in additional noise reduction because decimation calculates the mean value of consecutive samples, having the effect of low-pass filtering.

Stitching increases the conversion range of an ADC by executing one measurement with a low gain setting and a second measurement with a high gain setting in the amplifier path. The clipped part of the high-gain measurement is replaced by the corresponding detail of the nonclipped measurement multiplied by the ratio of the gain factors. The transition region of the resulting fiber trace is invisible to the user since the gain ratio is calculated from the two individual traces to eliminate any gain variation of the analog amplifiers and because the transition region is computed as a weighted moving average.