Conclusion: The results on a W-band monolithic varactor fre-
`quency doubler were reported for the first time, showing per-
`formance close to that predicted. The bandwidth of the
`
`92
`
`9 3
`94
`output frequency,GHz
`
`I
`95
`
`Fig. 3 Output power against outputfrequency of doubler
`8 predicted
`0 measured
`
`1
`16
`
`1
`18
`
`14
`
`1
`24
`
`1
`I
`20
`22
`2 6
`1639141
`input power,dBm
`Fig. 4 Measured output power and efficiency against input power of
`doubler at 93 GHz
`8 output power
`A efficiency
`doubler exceeds 4GHz. At 93GHz an output power of 30mW
`was obtained with a conversion elficiency of 12%. Further
`improvement in the conversion elficiency is expected by a
`reduction in the series resistance of the diode. These results
`establish the feasibility of generating W-band power using the
`monolithic doubler circuit approach. Higher output power
`can be achieved by connecting varactor diodes in series to
`increase the breakdown voltage.
`
`Acknowledgments: This work was supported by Hercules
`Defense Electronics Systems Inc. The authors would like to
`thank T. Smith, J. Singer, G. Tough, E. Chang and J. Kearney
`for their technical assistance, and Dr. H. Huang for his
`support.
`
`26th November 1990
`
`G. HEGAZI
`A. EZZEDDINE
`F. PHELLEPS
`P. McNALLY
`K. PANDE
`COMSAT Laboratories
`Clarksburg, MD 20871, USA
`P. RICE
`P. PAGES
`Hercules Defense Electronics Systems
`PO Box 4648, Clearwater, FL 33518, USA
`
`References
`1 METZE, G., LEE, T., BASS, I. F., LAUX, P. L., CARLSON, H. C., and CORN-
`FELD, A. 0 . : ‘A narrow-channel, 0.2flm gate-length, double-
`quantum-well pseudomorphic MODFET with high power gain at
`millimeter-wave frequencies’, to be published in IEEE Electron.
`Dev. Lett.
`
`21 4
`
`1 ~-
`
`2 MAIIDY-AHY, R., RIAZIAT, M., NISHIMOTO, C., GLENN, M., SILVERMAN,
`s., WENG, s., m A , Y. c., ZDASIAK, G., BANDY, s., and TAN, z.: ‘94GHz
`InP MMIC five-section distributed amplifier’, Electron. Lett., 1990,
`26, (2), pp. 91-92
`3 HEGAZI, C., WUNG, H.-L., SINGER, I. L., PHELLEPS, F. R., CORNFELD, A.
`B., SMITH, T., BASS, 1. F., CARLSON, H. E., and HUANG, H. C.: ‘GaAs
`MBE monolithic power amplifiers at U-band’. IEEE Microwave
`and Millimeter-Wave Monolithic Circuits Symp., Tech. Digest,
`1989, pp. 121-125
`
`4 CHU, A., COURTNEY, w. E., MAHONEY, L. I., MURPHY, R. A., and
`R. w.: ‘GaAs monolithic frequency doublers with
`MCLELLAND,
`series connected varactor diodes’. IEEE-MTT-S Symp. Digest,
`1984, pp. 51-54
`5 SIEGEL, P., KERR, A., and HWANG, W.: ‘Topics in the optimization Of
`millimeter-wave mixers’. NASA Technical papa 2287, March 1984
`
`AUTOMATIC ADJUSTMENT OF
`QUAD R ATU R E M 0 D U LATO RS
`
`Indexing terms: Modulators, Adaptive systems and control,
`Mobile radio svstems
`A technique for the automatic correction of errors in RF
`quadrature modulators is described. The method uses a
`diode detector to provide amplitude feedback to guide the
`adaption of baseband circuits that correct for carrier leak,
`differential gain and phase mismatch errors.
`
`Introduction: Modern digital mobile communications systems
`often use quadrature techniques to implement the desired
`modulation. The problems of quadrature circuits are well
`understood from early work in SSB and radar.’ The main
`problems are quadrature phase and differential gain errors
`between the in-phase ( I ) and the quadrature (Q) channels as
`well as carrier leakage. These parameters vary with tem-
`perature and applied carrier frequency f, often making
`readjustment necessary.
`
`Correction of errors: Adjustment of the baseband drive signals
`feeding the RF mixers can correct for these errors. The carrier
`leak can be compensated for by adding DC terms to the drive
`signals (i’ = i + i , and q’ = q + qoc. where i’, q’ are the pre-
`corrected mixer drive signals). The equations for the correc-
`tion of differential gain are i’ = iki and q’= qk,,
`and
`for correction of differential phase are
`
`i’ = i sec 4/2 - q tan 4/2 = sec 4 ~ 2 ( i - qk)
`q’ = q sec 4/2 - i tan 4/2 = sec 4/2(q - ik)
`where ki and k, adjust the I and Q gains and 4 is the quadra-
`ture phase error. The latter two equations separate the adjust-
`ment procedure into an axis twist component, k( = sin 4/2),
`and gain normalisation term, sec 4/2.
`It is possible to keep the corrections all independent of each
`other by applying the equations in the correct order (Fig. I).
`This allows the adjustment of one parameter without affecting
`the set up of the other two. Carrier leak compensation should
`be applied last so that any subsequent gain or phase adjust-
`ments will not modify the DC correction terms.
`Corrections for differential gain should be applied before
`the phase correction because the latter assumes that the gain
`
`I
`
`4
`
`Fig. 1 Signalflow graph ofIQ correction network
`ELECTRONICS LETTERS 31st January 1997 Vol. 27 No. 3
`
`~~~
`
`-.
`
`~
`
`SAMSUNG EXHIBIT 1012
`
`Page 1 of 3
`
`

`

`on both arms is matched while the former does not require an
`orthogonal axis. This decoupling greatly simplifies any algo-
`rithm that is used to adjust the correction network.
`
`the performance of the I , Q modulator.' The correction
`network and control algorithm were implemented on a digital
`0
`
`Adjustment: Some additional feedback is required for auto-
`matic adjustment of the modulator. By using the correct algo-
`rithm this can be limited to the output of a simple diode
`detector (Fig. 2). The algorithm proceeds in the following
`
`correction network
`
`,&
`
`if
`
`controller
`
`Fig. 2 Quadrature modulator and feedback
`manner. First the correction circuit is initialised (dci, dcq,
`6 = 0; ki, k, = 1) and then test vectors (Ai, A q ) are applied to
`the i and q inputs from which the necessary adjustments can
`be made. The adjustments are made from right to left on Fig.
`1.
`The carrier leak is corrected for by zeroing the drive signals
`i and q (test vector = 0, 0), and then adjusting the dCi and dcq
`terms to minimise the detector output. A simple and effective
`way to do this is to use the one dimensional search. In this
`search one term (for example d,)
`is incremented or decre-
`mented by a small step length while the other term (dei) is held
`constant. The change in output amplitude is recorded after
`each step. If the amplitude drops then the term is incremented
`(decremented) further, and this continues until the output
`amplitude of the modulator reaches a local minimum. The
`other direction (d,J is then selected for update. The process is
`repeated with reducing step lengths until an optimum is
`reached. For this algorithm the detector needs to have a high
`sensitivity and this is obtained by biasing the detector and
`providing amplification. There is no need for the detector to
`be linear or even calibrated provided it is monotonic and has
`a response that does not drift over the measurement period.
`The algorithm is robust in that it will still work in the pre-
`sence of considerable differential gain and phase errors.
`After the carrier leak has been nulled the measurement of
`gain along the two axes can proceed without being corrupted
`by offset errors. Test vectors (A, 0) and (0, A ) are separately
`applied and the amplitudes of the resulting outputs measured
`on the detector. The ratio of these two measurements is the
`gain mismatch, hut this is only accurate if the detector is
`calibrated. Alternatively, an iterative approach can avoid the
`requirement for a calibrated detector. In this approach the
`arm with the larger output has its gain (k, or k,) decremented
`until the outputs from both arms are equal.
`With the gain along each axis now equalised and the carrier
`leak nulled all that needs to be corrected is the phase. The
`phase error will produce an ellipse (instead of a circle) in the
`IQ plane which is centred at the origin when the system is fed
`with a constant amplitude rotating phasor [i = cos (2rrft) and
`q = sin (27~ft)l.~ If the angles between the I and Q axes are less
`than 90 degrees (Fig. 3) the maximum amplitude E,, occurs
`with the input vector (i = A, q = A), and the minimum E,,,
`with the vector (i = A, q = - A ) . Using co-ordinate geometry
`the phase error is related to E,( = EmX/EmiJ by (4/2) = arctan
`(E, - l/E, + 1).
`A method that avoids calculations iteratively adjusts the
`cross term k until the two amplitudes become equal (E, = 1.0).
`In the case of Fig. 3, k would be increased because E,,, occurs
`in the first quadrant. The gain terms (sec $42) are either incor-
`porated into other parts of the system (usually possible) or
`adjusted using sec 442 = [1/4(1 - k')] 1 1 + (k2/2) (accurate
`to 1 % for phase angle corrections up to
`= 45").
`
`Results: The algorithm has been successfully used for predis-
`tort linearisation circuits which are particularly sensitive to
`
`Fig. 3 Quadrature phase error
`
`signal processing device. The 900 MHz quadrature modulator
`consisted of a 90" phase splitter, a combiner and two diode
`ring mixers fed at a + 7 dBm RF drive level. These units were
`all commercially available 'building block' products. No par-
`ticular care was taken to keep lead lengths even. The uncor-
`rected modulator had an overall differential phase error of 8",
`a differential gain of O.SdB, and carrier leak of -20dBm.
`After correction the above errors improved to O@, 0,02dB,
`and -63dBm, respectively, and a plot of a constant ampli-
`tude rotating phasor, as recorded on a network analyser, is
`shown in Fig. 4. The magnitude variation due to all error
`sources is 307.31 pU (0.065dB). Noise and mixer nonlinearity
`were the major source of adjustment error.
`
`LK+ GAIN+PHASE
`2 307 31 p U - 1 3 6 79'
`h R E F = l
`
`.full
`
`Fig. 4 Rotating phasor after correction, gain phase plot (1°C'
`scale)
`Conclusion: The algorithm gives repeatable results, requires a
`simple diode detector for the feedback signal and its iterative
`nature avoids the need for any calculations. This makes it
`suitable for implementation on low cost controllers. Further-
`more, it is possible to modify the algorithm to avoid the need
`for special test vectors by sampling the detector at the correct
`point in the trajectory of the incoming modulation. In this
`case it is more suitable to adjust the carrier leak when the
`trajectory crosses the I or Q axis at a, b, c and d (Fig. 3) and
`then adjust the DC terms to make the amplitudes a = c and
`b = d . Modulations with quadrature amplitude symmetry are
`most suitable, e.q. QPSK, CPM etc.
`
`M. FAULKNER
`6th November 1990
`Department of Electrical and Electronic Engineering
`Victoria University of Technology (Footscray campus)
`Melbourne, Australia
`T. MATTSSON
`Department of Appl~ed Electronics
`Lund Uniuersity
`Lund, Sweden
`W. YATES
`Department of Electrical Engineering
`University of Technology Sydney
`Sydney, Australia
`ELECTRONICS LETTERS 31st January 1991 Vol 27 No 3
`
`- -1
`
`21 5
`
`Page 2 of 3
`
`

`

`Referenees
`1 SINSKY, A. I., and WANG, P. c. P.: ‘Error analysis of a quadrature
`coherent detector processor’, IEEE Trans., 1974, AES, pp.
`88C883
`2 ROOME, s. I.: ‘Analysis of quadrature detectors using complex
`envelope notation’, IEE Proc. F, 1989,136, (2), pp. 95-100
`3 FAULKNER, M., MATWON, T., and YATES, w.: ‘Adaptive linearisation
`using predistortion’. Proceedings of the 40th IEEE Vehicular
`Technology conference, 1990
`
`CASCADABLE LASER LOGIC DEVICES:
`DISCRETE INTEGRATION OF
`PHOTOTRANSISTORS WITH
`SURFACE- EMITTING LASER DlOD ES
`
`Indexing t e r m : Logic devices, Optical logic
`A new class of cascadable optical logic devices is described
`which consists of integrated GaAsIAIGaAs phototransistors
`and surfaceemitting lasers. The devices function as optical
`neurons (‘smart pixels.), have high optical gain (a factor of
`2 2 0 overall, 2200 differential), very high on/off contrast
`(>2700, 34dB) and are ideal for neural networks and optical
`computing applications.
`
`We describe a cascadable optical switch based on the integra-
`tion of a heterojunction phototransistor (HPT) with a vertical-
`cavity surface-emitting laser (VCSEL) diode, which we call a
`surfaceemitting laser logic (CELL) device.’ We have demon-
`strated high optical gain (a factor of >20 overall, 2200
`differential) with the discrete but integrable components con-
`nected in series. Our devices are ideally suited for parallel
`optical signal processing which has been severely hampered
`by the lack of several basic ‘building blocks’ such as optical
`switching devices that exhibit low switching energy, high
`optical gain (fanout capability), cascadability, and high con-
`trast. Ideal devices should also be easy to fabricate (use self-
`aligned VLSI
`technology), microscopic
`in size, readily
`integrable with other optoelectronic components without
`being adversely affected by external optical feedback, tolerant
`to temperature variations, and require a minimum number of
`components.
`Whereas HPTs and VCSELs have been demonstrated in
`previous work,’,’ here we describe the properly scaled, inte-
`grable versions of these components and the hybrid operation
`of the pair. In the operation of the CELL device, incident light
`generates a photocurrent in the detector element which is
`internally amplified and then used to drive the laser element
`above threshold. Therefore each CELL acts as an independent
`thresholding optical amplifier. This optical-electrical-optical
`conversion results in a robust high-gain device in which the
`input optical beam is unaffected by external optical feedback.
`Due to the large absorption bandwidth of the HPT, the
`devices are able to convert beams from one wavelength to
`another. They can also convert incoherent light to coherent
`light. The devices are easily fabricated into high-density, 2-D
`arrays. As both the input and output images are directed
`perpendicular to the arrays, CELLs will be extremely useful
`for parallel signal processing, multichannel interconnections,
`neural networks and visual displays. CELLs have an impor-
`tant advantage over photothyristor devices. Whereas those
`devices are latching and need to be reset electrically, CELLs
`can be designed not to latch. CELLs can also be configured as
`latching devices, in which case they can function as switching
`arrays for optical symbolic substitution or as image memories
`capable of capturing, storing and displaying a 2-D image until
`arrays are reset.
`In order to achieve a high-gain optical device that is cas-
`cadable and insensitive to feedback, we use the light actuated
`current switching capability of a high-gain AlGaAs/GaAs
`HPT and a low-threshold, high-power AlGaAs/GaAs VCSEL.
`
`The npn HPT, grown by MBE, consists of a wide-band-gap
`(3000A, n = 1 x 1017m-3) and
`A1,.,,Ga0.,,As
`emitter
`narrow-band-gap GaAs base (2350A, p = 1 x lO’*~m-~)),
`(3000A, n = 1 x 10t6cm-’),
`collector
`and
`subcollector
`(1670A, n = 1 x 10’8cm-3) regions. The wide-band-gap
`emitter inhibits the base-to-emitter hole injection current and
`significantly increases the HPT gain. We electrically isolated
`the HPTs using shallow proton implants. We employed a
`bilayer structure consisting of a layer of electroplated gold
`covering a layer of photoresist as an ion blocking mask and
`also as a liftoff mask to form selfaligned electrical contacts.
`The input apertures of the HPTs have 15pm diameters and
`the implant-defined device size is 40 x 40pm. The photo-
`transistor
`common-emitter,
`floating-base
`characteristic
`exhibits
`lOmA photogenerated current
`for 120pW of
`absorbed light power (176pW incident) at 800nm. The MBE
`20
`of
`grown VCSELs
`consist
`a
`period
`p-type
`AIAs/Al,.,,Ga,.,,As distributed Bragg reflector (DBR), a four
`quantum-well (100 A) GaAs/AI,.,Ga,.,As
`active region and a
`27.5 period n-type AI,.,,Ga,.,,As/AIAs DBR.’ We electrically
`isolated the VCSELs in a manner similar to that used for the
`HPTs by using deep-proton implants3 with 15pm diameter
`implant masks and selfaligned contacts. The CW room-
`temperature laser diode L-I characteristic exhibits 2.5 mW
`laser-light output power at 850 nm for 10 mA injection current
`and is therefore ideally suited for integration with the ion
`implanted HPT.
`In Fig. 1 we show the calculated energy diagram and the
`equivalent circuit for a CELL. R, and R, refer to the series
`resistance of the n-type and p-type mirrors, respectively. R,
`dominates the series resistance of the VCSEL4 resulting in
`nonideal laser diode behaviour. The large R, is due to voltage
`
`ORn
`
`h”out
`
`n mirror
`AlGoAsI A l A s
`
`subcollector
`
`vaiCnce
`band
`
`conduct ton
`band
`
`fjgh”\n E”
`-
`
`A
`Fig. 1A Calculated energy-band diagram ofintegrated HPTIVCSEL
`Diagram is to scale with exception that the 4-QW active region is
`expanded five times; only the lowest-energy conduction band is
`shown
`Fig. 16 Equiualent circuit 0fCEL.L
`R, = resistance of n mirror
`R, = resistance of p mirror
`V, = bias voltage
`
`spikes at the AIAs/AlGaAs valence-band interfaces arising
`from equilibrium charge transfer. The voltage spikes are
`apparent in the energy diagram shown in Fig. l a . In practice
`we incorporate thin AI,.,,Ga,.,,As
`steps at each interface of
`the p-type DBR to reduce this resistance to a level on the
`order of the resistance of the n-type mirror.
`In Fig. 2 we show the VCSEL diode output power as a
`function of the input power to the HPT. These transfer char-
`acteristics can be modified by adjusting the supply voltage.
`The CELL has both high overall optical gain (a factor of >20
`
`21 6
`
`-1
`
`ELECTRONICS LETTERS 31st January 1991 Vol. 27 No. 3
`
`Page 3 of 3
`
`