Researchers at the University of California, Los Angeles (UCLA) have found an optical way to speed up the analog-to-digital (A/D) conversion process that currently limits the use of available digital-signal-processing capacity for high-performance communication and radar systems. The new method is based on technology that has already been developed for telecommunications, particularly in the area of wavelength-division multiplexing, but requires better signal-to-noise performance than do telecom
Optics circumvent bottleneck in A/D conversion
Researchers at the University of California, Los Angeles (UCLA) have found an optical way to speed up the analog-to-digital (A/D) conversion process that currently limits the use of available digital-signal-processing capacity for high-performance communication and radar systems. The new method is based on technology that has already been developed for telecommunications, particularly in the area of wavelength-division multiplexing, but requires better signal-to-noise performance than do telecom applications.
A wide range of defense and telecom applications stand to benefit from development of the UCLA method. The Defense Advanced Research Projects Agency (DARPA) has provided multimillion-dollar funding for a four-year, multi-institution consortium led by UCLA to build a 40-Gsample/s, 8-bit, A/D converter.
The original problem might be thought of as a need to balance the contradictory effects of Moore`s Law and Murphy`s Law. The doubling every 18 months of chip capacities described in Moore`s Law has currently driven digital-signal-processor performance to several giga operations per second, according to Bahram Jalali, the principal investigator on the project.
Such fast system speeds offer tremendous opportunities for high-performance digital-signal processing in applications ranging from radar and electronic warfare to cellular telephony. But as Murphy`s Law would have it, a bottleneck has arisen in the mature technologies used for A/D conversion. Instead of doubling annually, the rate of improvement for a given speed is only about 1 bit every five years.
"The fastest analog-to-digital converter that I can buy with 10-bit resolution is about 100 Mbit/s," Jalali said. "If you compare that to several giga operations per second of processing power, there is a huge gap. As a result, the A/D converter is the bottleneck in system performance."
Because it didn`t seem likely that A/D processing speeds would catch up with PCs anytime soon, Jalali`s group took another approach. "We said, `Instead of making the A/D converter faster, can we make the [analog] signal slower?`" If so, the high-speed analog signal could be segmented and interleaved into parallel channels where each channel could be slowed down enough for A/D conversion with available technology without losing any information from the original signal.
The UCLA researchers came up with a method of time-stretching incoming analog signals based on work that had already been done in the optical community on stretching and compression of Gaussian signals. The method developed for optical pulses involved a three-step process of dispersing the signal, chirping it, and dispersing it again. To apply this technique to real-time signals, the UCLA team came up with the segmenting and interleaving scheme mentioned above and also observed that the signal-stretching process could be simplified by using a chirp bandwidth much larger than the bandwidth of the signal to be stretched.
"If I use a very short [femtosecond-width] optical pulse, then I can obtain a bandwidth of several hundred terahertz, whereas the signal I am trying to stretch is, at most, several hundred gigahertz," Jalali said. "So by performing a chirp using ultrashort pulses, we can satisfy the condition [of a much higher chirp bandwidth]. That gives us a very simple system that we can implement using commercially available components."
The wide-bandwidth (7.5 THz) optical chirp pulse for the time-stretching technique is generated by passing a 160-fs pulse from a modelocked, erbium-doped fiber ring laser through 1.1 km of single-mode optical fiber. Intensity modulation of the chirp pulse by the analog input signal is performed in an electro-optic (lithium niobate) modulator. The signal is then time-stretched by passing through a second stage of single-mode optical fiber of 7.6 km in length (see figure).2
1. A. S. Buhshan, F. Coppinger and B. Jalali, Electron. Lett. 34(9), 839 (April 30, 1998).
2. B. Jalali, A. S. Bhushan, and F. Coppinger, "Photonic Time-Stretch: A Potential Solution for Ultrafast A/D Conversion," IEEE Microwave Photonic Conference, MWP `98, Princeton, NJ (October 1998).