September 15, 1995 / Vol. 20, No. 18 / OPTICS LETTERS 1901

Optical folding-flash analog-to-digitalconverter with analog encoding

B. Jalali and Y. M. Xie

Department of Electrical Engineering, University of California, Los Angeles, Los Angeles, California 90024-1594

Received May 1, 1995

We describe an optically assisted folding-f lash analog-to-digital converter. The periodic transfer function ofthe Mach–Zehnder interferometer is used to perform analog folding on the electronic signal to be quantized.A novel analog encoding scheme for efficient generation of gray code digital data is proposed. The newencoding scheme eliminates the requirement for interferometers with ultralow Vp , which, so far, has hinderedthe development of such systems. The encoding concept is experimentally demonstrated through the use ofLiNbO3 modulators. 1995 Optical Society of America

The development of high-resolution, high-sampling-rate analog-to-digital converters (ADC’s) is imperativebecause of the growing number of applications that usehigh-speed digital signal processing, such as wirelesscommunication, radar, and electronic warfare. Thetrend in modern communication systems is to usedirect receivers, thereby reducing the number ofdownconversion steps. These systems require analog-to-digital (AyD) conversion at rf frequencies. Anexample is the electronic warfare system in which anincoming radar signal is digitized at rf frequenciesand stored in a high-speed digital rf memory. Whenthese digital data are then converted back to analogform and transmitted to the originator, their signalprocessing algorithms can be confused. Such applica-tions place stringent requirements on the bandwidthof ADC’s, which, in many cases, is beyond the reach ofelectronic circuits.

In this Letter we propose an ADC that uses opticalfolding and analog encoding to perform electronic AyDconversion at ultrahigh frequencies. The converteris based on Mach–Zehnder (MZ) interferometers andcan be fully integrated on a wafer. The encodingtechnique eliminates the requirement for ultralowVp , which, so far, has hindered the development ofsuch systems. To illustrate the proposed architecture,operation of a 4-bit ADC is described.

The use of a MZ interferometer for AyD conversionwas first suggested by Taylor,1 who recognized the con-nection between the electro-optical transfer function ofthe device and the gray-code digital data. This con-cept is shown in Fig. 1. The analog electronic signal issimultaneously applied to an array of interferometersconfigured in parallel. The sinusoidal transfer func-tion of the MZ interferometer is used to fold the analoginput signal symmetrically. Each MZ interferometerhas an active length that is twice that of its nearestmore significant bit and hence has twice the folding fre-quency. The resulting optical waveform is a gray-coderepresentation of the analog input. Digital electronicdata are generated with high-speed photodetectors andcomparators. This scheme has many highly attrac-tive features, including (i) possibility of AyD conversionat millimeter-wave frequencies, (ii) direct encoding,

0146-9592/95/181901-03$6.00/0

(iii) ultrafast low-jitter sampling, and (iv) isolation ofthe input and sampling signals.

Based on this concept, a number of prototype electro-optic ADC’s have been demonstrated.2 – 4 Althoughthe ADC concept shown in Fig. 1 is simple, it hasa major drawback that makes it impractical. In aMZ interferometer the output intensity Io varies withanalog voltage Va as5

IosVad ­ Ii cos2µ

pVa

2Vp

∂, (1)

where Ii is the input intensity and Vp is the half-wavevoltage. To generate the least significant bit (LSB) inan N -bit quantizer, the Vp of the modulator must scaleas

Vp ­ s1y2N dVfs , (2)

where Vfs is the full-scale input analog signal. Therequired geometrical scaling of Vp with resolutionrenders this technique impractical for ADC’s of evenmodest resolution. For example, a 4-bit ADC withVfs ­ 10 V will require Vp ­ 630 mV, which is beyond

Fig. 1. Direct conversion of an analog signal into graycode through an array of MZ interferometers with binary-scaled active lengths.

1995 Optical Society of America

1902 OPTICS LETTERS / Vol. 20, No. 18 / September 15, 1995

what can be achieved with state-of-the-art MZ interfer-ometers based on either LiNbO3 (Ref. 6) or electro-opticpolymers.7

To extend the resolution of interferometer-basedADC’s, Pace and Styer proposed using the symmetricalnumber system.8 This technique extends the resolu-tion by providing more than one bit per interferometer.It significantly reduces the number of MZ interfer-ometers required for a given resolution; however, asmall electrode length difference (or a small signal at-tenuation) is necessary for each interferometer. Fur-thermore, unlike with Taylor’s design, a separatethermometer-code-to-binary-encoding circuitry in nec-essary, and electronic comparators require nonuniformand complex threshold settings.

It is possible to eliminate geometrical scaling of Vp

with the resolution described by Eq. (2) and to main-tain the efficiency and simplicity of architecture shownin Fig. 1 by incorporation of analog encoding. To illus-trate the concept, we describe the operation of a 4-bitfolding-f lash ADC, shown in Fig. 2. In this schemethe most signif icant bit (MSB) is generated by a MZinterferometer with Vp ­ Vfsy2 and biased at point Bon the transfer characteristics. The second bit is ob-tained by a MZ interferometer with the same activelength (same Vp ) but biased at point A. This ensurestwo zero crossings within the full-scale range, as shownby the output waveforms in Fig. 2(b). The remainingbits are generated by encoding folded waveforms of ap-propriate phase generated by interferometers identicalto those used in the first two bits.

The phase of the folded waveform can be controlledthrough the dc bias point of the interferometer. Forexample, when the interferometer is biased at point C0,its waveform is described by

IC0 ­Ii

2

∑1 2 cos

µpVa

Vp

∂∏(3)

and is 180± out of phase with that at bias pointA. When the waveforms corresponding to bias pointsA [Eq. (1)] and C0 are multiplied in the voltage domain,the resulting optical intensity has twice the foldingfrequency:

IMSB22 ­ IAIC0 ­Ii

8

∑1 2 cos

µ2pVa

Vp

∂∏. (4)

Multiplication in the optical domain can be performedby series connection of two interferometers, as shownin Fig. 2(b). The resulting waveform, displayed inFig. 2(c), has the precise phase relation with respectto MSB and MSB 2 1 to represent the MSB 2 2 bit.

The LSB has twice the folding frequency of theMSB 2 2 bit and is generated by multiplication of fourphase-shift waveforms corresponding to bias points A,C0, B, and B0:

ILSB ­ IAIC0IBIB0 ­Ii

128

∑1 2 cos

µ4pVa

Vp

∂∏. (5)

This has the correct folding frequency and phase torepresent the LSB.

Figure 2(c) shows the simulated waveforms of theproposed ADC obtained by the simulation program

with integrated circuit emphasis (SPICE) circuit simu-lation tool. The MZ modulator was modeled in SPICEby analog behavioral modeling, and a dc sweep is per-formed over an input voltage range Vfs ­ 2Vp . Asexpected, the output of the MZ interferometer is a4-bit gray code representation of the analog signal.The MSB is potentially susceptible to noise for extremevalues of the input voltage range sVa ­ 6Vp d, whichone can easily eliminate by providing an arbitrary andsmall attenuation (by a resistor) of the input signal thatdrives the MSB modulator.

Figure 3 shows experimental data derived from twoUnited Technologies LiNbO3 MZ modulators config-ured in series. The modulators have a Vp of 5.3 Vfor l ­ 1.55 mm. The MSB and MSB 2 1 bits rep-resent the output of a single modulator biased at22.65 V (point B) and 0 V (point A), respectively. TheMSB 2 2 bit is generated by application of the full-scale analog signal simultaneously across two modula-tors biased at 0 V and 25.3 V. The amplitude of the

Fig. 2. (a) Transfer characteristics of the MZ interferome-ter, (b) block diagram of the proposed ADC, (c) output show-ing the digital gray-code representation of the analog input.

September 15, 1995 / Vol. 20, No. 18 / OPTICS LETTERS 1903

Fig. 3. Experimental data derived from two series-connected LiNbO3 MZ modulators.

MSB 2 2 bit is scaled to facilitate the comparison of thethreshold crossings. The data represent a 3-bit gray-code representation of the analog signal.

The proposed ADC eliminates the geometrical scal-ing of Vp required in the Taylor scheme (Fig. 1). Forexample, for the 4-bit ADC the conventional architec-ture of Fig. 1 would require a MZ interferometer withVp ­ Vfsy16, compared with Vfsy2 for the new architec-ture. This eliminates the requirement for modulatorswith ultralow Vp and is critical for the realizationof practical ADC’s. This improvement is obtainedat the expense of using only one additional interfer-ometer. Also, all interferometers in the ADC have thesame electrode length, a property that is importantfor monolithic integration with integrated waveguidemodulators. Unintentional differences in the Vp ofconstituent devices (caused by processing-inducedvariations), for example, can be compensated for by thebias point setting of individual modulators and by ap-propriate scaling of Vfs, when resistors are used in thesignal path.

Compared with an electronic f lash converter, theoptical ADC is highly efficient because it generatesthe gray-code digital data without a separate encod-ing process. The only electronic components requiredare a photodetector, a transimpedance amplifier, andlatched comparators (decision circuits), all of which canoperate at millimeter-wave frequencies.

Higher-resolution converters can be constructedin a similar manner. For example, the next LSBcan be generated by multiplication of phase-shifted

folded waveforms that correspond to bias pointsAC0B0BDD0EE0. In general, the number of interfer-ometers in the ADC scales as 2sN22d 1 1, where N isthe desired resolution. Another potential limitationof the proposed architecture is the scaling of theamplitude with higher-order bits. Amplifiers (opticalor electronic) may be used to compensate for this. Ingeneral, direct extension of this architecture to con-verters with more than 6 bits may not be practical.For ADC’s with moderate resolution, the optimum con-figuration will be a hybrid approach. For example, anultrafast 8-bit ADC can be realized with a 4-bit opticalfolding–encoding processor to extend the resolu-tion of a high-speed 4-bit electronic f lash converters.In general, performing the folding function opticallyresults in improvement in the performance of folding-f lash circuits because their bandwidth is largelylimited by that of the folding block.

In summary, we have demonstrated an opticalfolding-f lash AyD converter that uses a MZ interfer-ometer to perform folding and analog encoding. Theproposed encoding scheme eliminates the requiredgeometrical scaling of the modulator Vp with AyDresolution.

The authors thank H. R. Fetterman for helpfuldiscussions.

References

1. H. F. Taylor, IEEE J. Quantum Electron. QE-15, 210(1979).

2. G. D. H. King and R. Cebulski, Electron. Lett. 18, 1099(1982).

3. R. A. Becker, E. E. Woodward, F. J. Leonberger, andR. C. Williamson, Proc. IEEE 72, 802 (1984).

4. R. G. Walker, I. Bennion, and A. C. Carter, Electron.Lett. 25, 1443 (1989).

5. B. E. A. Saleh and M. C. Teich, Fundamentals ofPhotonics (Wiley, New York, 1991), Chap. 18, p. 704.

6. See ‘‘Specif ication for LiNbO3 modulators from UnitedTechnologies Photonics’’ (United Technologies, 1289Blue Hills Avenue, Bloomfield, Conn. 06002).

7. W. Wang, D. Chen, and H. R. Fetterman, Appl. Phys.Lett. 65, 929 (1994).

8. P. E. Pace and D. Styer, Opt. Eng. 33, 2638 (1994).