N92-14235

FLEXIBLE DIGITAL MODULATION AND CODING SYNTHESISFOR SATELLITE COMMUNICATIONS

Mark Vanderaar*

Sverdrup Technology, Inc.2001 Aerospace Parkway

Brookpark, Ohio 44142

James BudingerNASA Lewis Research Center

Mail Stop 54-821000 Brookpark Rd.

Cleveland, Ohio 44135

Craig Hoerig and Dr. John Tague**Dept. of Electrical and Computer

EngineeringOhio University

Athens, Ohio 45701

ABSTRACT

This paper presents an architecture and a hardwareprototype of a Flexible Trellis Modem / Codec (FTMC)transmitter. The theory of operation is built upon a pragmaticapproach to trellis-coded modulation that emphasizes powerand spectral efficiency [1]. The system incorporatesprogrammable modulation formats, variations of trellis-coding, digital baseband pulse-shaping, and digital channelprecompensation. The modulation formats examined include(uncoded and coded) Binary Phase Shift Keying (BPSK),Quartenary Phase Shift Keying (QPSK) , Octal Phase ShiftKeying (8PSK), 16-ary Quadrature Amplitude Modulation (16-QAM), and Quadrature Quadrature Phase Shift Keying

(Q2pSK) at programmable rates up to 20 Megabits persecond (Mbps). The FTMC is part of a developing test bed toquantify modulation and coding concepts.

INTRODUCTION

Regenerative transponders employing onboard-processingand switching techniques are active topics in satellitecommunications [2, 3, 4]. One common feature among thearchitectures proposed is the presence of modems andcodecs both on the ground and in the spacecraft. Themodems and codecs are used with multiple access schemesdesigned to optimize both uplink and downlink capacities.Generally, they trade some mix of bandwidth and powerefficiency.

Frequency Division Multiple Access (FDMA) is an appropriatemultiple access technique for low-rate bandwidth efficientuplinks from a large number of ground terminals. In this typeof uplink, bandwidth efficiency is directly related to theamount of uplink traffic attainable. Time Division Multiplexing(TDM) is envisioned in high-rate downlinks for its powerefficiency. In TDM, nonlinear High Power Amplifiers (HPA's)can be operated at saturation without the adverse effect ofintermodulation and distortion as long as the modulatedsignal maintains a constant envelope. Spread spectrumtechniques such as Code Division Multiple Access (CDMA)are used for their power efficiency for the same reasons asTDM. CDMA has also shown increased spectral efficiency insome systems and may be suitable for both uplinks anddownlinks.

In the future, as data throughput requirements increase tooffer wider band services, TDM or CDM downlinks may notonly be limited by available transmit power but also bychannel, transponder, and ground terminal bandwidths. For

*Work Supported by Contract NAS3-25266, Task Order5601, Digital Systems Technology Development.

**Work Supported by Grant NAG3-1183, High PrecisionWaveform Equalization for Optimum Digital Signalling.

example, Travelling Wave Tube Amplifiers (TWTAs) qualifiedfor space environments today exhibit bandwidths on theorder of 1 to 5 percent of the operating frequency [5]. Higheroutput power TWTAs generally afford lower bandwidthpercentages. Technologies such as electronically steerablephased arrays that employ patch antennas can be designedto exhibit bandwidths on the order of 10 percent of theiroperating frequencies, though they become simpler andsmaller at more modest bandwidths of 1 to 3 percent. Also,digital modem and codec hardware can be designed tooperate almost exclusively at the symbol rate, as opposed tothe bit rate. Thus, higher bit throughput rates can beaccomplished at lower processing rate by using modulationschemes with more bits per symbol. This may be useful in aTDM downlink exhibiting high rates to a number of low-costground terminals.

Increased flexibility is another dimension in future satellitesystems. The on-orbit reconfiguration of the modulation andcoding hardware offers the potential to provide more networkcapacity, increased network availability, multiple serviceclasses, and higher percentage availability at the expense ofincreased complexity. Efficient use of capacity could beobtained with a network that charges users for the amount ofnetwork resources required. Users that require high data andlow error rate services would expend more channelbandwidth and/or power per transmitted bit than user withlower quality and rate service requirements. Wideravailability may be obtained with flexible transmissionformats suited to particular user requirements. Higherpercentage availability may be achieved by reconfiguring tomore conservative modulation and coding formatsovercoming outages such as rain attenuation and multipathfading at the expense of network capacity.

This set of possible operating environments warrant theinvestigation of modems and codecs that can offer futuresatellite systems increased operational performance andflexibility.

The first section of this paper presents an overview of themodulation and coding test bed under development. Thesecond section describes the modulation and codingformats supported by the FTMC. The next section developsthe digital pulse shaping techniques. Digital basebandprecompensation methods for nonlinear channel distortionare then described. The prototype transmitter that includesthe baseband modulator and IF upconversion is describednext. Finally, conclusions are drawn as to the applicabilitythe techniques presented.

SYSTEM CONFIGURATION

Though this paper focuses on the .FTMC transmitter, a briefdescription is given of the test bed under development. The

295

https://ntrs.nasa.gov/search.jsp?R=19920005017 2020-05-10T11:00:48+00:00Z

proposed FTMC test bed architecture is shown in Figure 1.System control is performed through a personal computerthat features a graphical user interface for ease of systemconfiguration and operation. FTMC configuration data isgenerated via object-oriented simulation software on aworkstation. The real time data flow through the FTMCbegins with either Pseudo-Random Binary Sequence (PRBS)or patterned test data generated by a digital transmissionanalyzer (DTA). The DTA is clocked with the FTMC modulatorhardware at the programmed data rate. TDM formatting isdone by the burst controller. The baseband modulatedwaveforrn is then generated by the FTMC modulator.Translation to a 70 MHz IF is performed through quadratureupconversion techniques. The data can then either betranslated to RF for amplification via a TWTA or directly to acalibrated IF noise combiner. The signal is thendownconverted to baseband where it is filtered,demodulated, and decoded by the FTMC demodulatorcurrently under development.

m l(t ) = h(t) * sl(t )

mQ(t) = h(t) * SQ(t)

where Sl(t ) and SQ(t) are the information carrying signals.The information carrying signals can be represented assquare pulses over the symbol time Ts of an amplitudecorresponding to the constellation definition. The functionh(t) is the impulse response of the bandlimiting pulse shapeand is described in the next section. Convolution is denoted

by the * symbol. In the case of Q2pSK the information

carrying signals correspond to the following expressions

Sl(t) = al(t)*pl(t) + a2(t)*p2(t)

SQ(t) = a3(t)*p1(t) + a4(t)*p2(t)

PC Compatible _ I I VAXstationCo_U_ I_i I_ II/GPX

RF Bypass Noise Source

Digital ModulatorAnalog IF RF Translation / Noise

Upconversion TWTA Combiner

Amplifier

Digital TransmissionAnalyzer

Analog IFDownconversion

FTMC

Digital Demodulator

Figure 1. FTMC Test Bed

MODULATION AND CODINGThe FTMC supports a number of different modulationformats. Each can be represented in quadrature form by theequation

Xtx(t) = ml(t)cos(2_fct) + mQ(t)sin(2_fct)

where each an(t ) corresponds to a positive or negativeamplitude, depending on the information sequence. The Pl (t)and P2(t) are a set of two orthogonal pulse shapes that alsomaintain orthogonality with the quadrature mixing operation.These conditions can be defined explicitly by the relations

where Xtx(t) denotes the resulting modulation at a carrierfrequency fc. The signals ml(t ) and mQ(t) represent thebaseband modulating functions. Three Phase Shift Keyed(PSK) techniques are supported that include BPSK, QPSK,and 8PSK. An amplitude and phase modulated technique,16-QAM, as well as a phase and pulse-shape modulatedtechnique Q2pSK [6], are also supported . In the PSK andQAM cases the baseband modulating functions representthe in-phase and quadrature bandlimited values associatedwith the signalling constellation. These can be written as

In n+ p,(t)ps(t )dt = 0

)Ts1

Ts

(n+l)TSp_t)eJ2_Ctdt = 0, i=1,2Ts

296

,c m=,ntogor_ lwhere n is an integer to define integration over any symboltime. There are an infinite number of pi(t)'s that satisfy the

above conditions. A few are developed in [6] that addressthe issue of phase continuity as well as spectral occupancy.In general, any modulation format whose Sl(t) and SQ(t) canbe represented by two or fewer bits can be created by theFTMC modulator. This is clarified in the waveform synthesissection.

The bit to symbol mapping of all the modulation formats aredesigned to operate in conjunction with a standardconvolutional code of constraint length K=7 and rate 1/2 withgenerator polynomials G 11=1338 and G12=1718 exhibiting

the maximal attainable dfree of 10 (Figure 2). This approachallows the use of a single encoder and decoder to be usedwith the programmable modulation formats to achieve a 3 to 7dB coding gains at real spectral efficiencies (using 40%Nyquist pulse-shaping filters) up to 2.14 bits/second/Hz.Coding gain can be approximated accurately at lower bit-error-rates by the Asymptotic Coding Gain (ACG). Thecalculation of ACG is based on the most likely decoding errorevent. "/'he most likely decoding error event for unmergedpaths with length greater than one is called the UnmergedSquared Euclidean Distance (USED). As shown in [I] USEDfor MPSK can be expressed as

+ ( dfree- 4 )sin 2 = M>4

where M is the modulation order of the coded scheme.Codes with parallel branches (and thus single branch errorsevents) have a Parallel Squared Euclidean Distance (PSED),again for MSPK, expressed as

M_>8

ACG is defined as the ratio of the minimum of the USED orPSED to the Normalized Euclidean Distance (NED) foruncoded performance. NED can be written as

and thus the ACG (in dB) for MPSK is

Evaluations of this expression are given in Table 1. Uncodedperformance was calculated using standard bounds [7, 8] forgray coded mappings. Note that for M=4 ACG is determinedby the USED and for M=8 ACG is determined by the PSED. Asimilar argument will be discussed to evaluate QAMperformance.

Modulation Code Asymptotic

Format Rate(s) Coding Gain

BPSK NA NAQPSK NA NA8PSK NA NA16QAM NA NATQPSK 1/2 7.0 dBT8PSK 2/3 3.0 dBT16QAM 2/4 4.3 dBT16QAM 3/4 3.6 dB

~Eb/No forPb = 10 -6

10.5 dB10.5 dB14.0 dB14.4 dB4.5 dB

7.5 dB6.5 dB10.4 dB

SpectralEfficiency*

0.721.432.142.86O.721.431.432.14

In bits/sec/Hz assuming 40% Nyquist Filtering

Table 1 : Modulation and Coding Performance

Gll

_ G12

Figure 2 : Rate 1/2, k=7 Convolutional Encoder

The particulars of each coding scheme can be described bya generator matrix of the form

Gn_,k =

G11 G12 ..- Gtk

-G21 - • G2k

: ". ".. !

where n is the number of input bits and k is the number ofoutput bits from the encoder. The Gij's are the activatedconvolutional encoder shift register taps written in an octalform. The k output bits are mapped into a 2k- ary modulationscheme in a manner that maximizes the Euclidean distanceof the code. For trellis-coded QPSK (TQPSK) the generatormatrix is

G1-_2=_ 1338 171e ]

Each information bit is coded into one QPSK symbol. A graymapping of the encoded bits is used. For this code the ACGis determined by the USED of the seven symbol long errorevent path. The code and mapping structure is illustrated inFigure 3. Note the encoder output bits read from the top tobottom correspond to the bit to symbol assignments readfrom left to right.

297

bl,bObl

Figure 3. Trellis Coded QPSK

Similarly, for T8PSK and T16QAM the generator matrices are

G2-,3=[ 133e171e 0 ]0 0 1008

1338 1718 0 0 JG3--,4 = 0 0 100e 0

0 0 0 1008

The code and mapping structures are shown in Figures 4 and5.

b2, bl, b¢b2

Figure 4: Trellis Coded 8PSK

011 001

"1"010• e00C

10o _10

lol 111

,,=)

b3, b2, bl, b(

bO

Figure 5 :Trellis Coded 16-QAM (Mapping #1)

Additionally, a rate 2/4 time-varying code is supported for the16QAM modulation format. This code maps two informationbits into one of 16 modulation points. The first bit is encodedinto a two bit symbol that determines the quadrant of the 16-ary symbol. The second information bit is encoded into a twobit symbol that chooses a "sector" within the previouslychosen quadrant. This sector and quadrant determine aconstellation point for the 16QAM format. The generatormatrix is written as

10o0 1101 110o 1001B • • •

1111 1010 1011 1110

010o 00Ol oooo 0101

0Oll o110 0111 0OlO

1338 171 1338 1718]

where cdenotes a demultiplexing operation. Theconstellation mapping and the code structure is illustrated inFigure 6.

b3, b2, bl, I_

1>3

b2

bO

o10o• 0101•]oo_ol o0e0o

1110• 1111•11011• 1010•1100 1101 "1001 1000

Figure 6: Trellis Coded 16 QAM (Mapping #2)

For Mapping #1 the PSED is the dominated error event.However, it is difficult to calculate due to the nature of the

decision regions. However, its performance is given in [1] asapproximately 10.4 dB Eb/No required to obtain a bit-error-

rate of 10 "6. This is a coding gain of 3.6 dB as compared to

gray coded 8PSK.

For Mapping #2 there are no parallel paths so the PSED is notan issue. The approximate USED is 1.34 as calculated bycomparing the first error event path with the all zero path inthe trellis diagram. The ACG as compared to the NED ofQPSK is

, o 0o,0i l oo,0r,L t4

WAVEFORM SYNTHESIS

To reduce spectral occupancy, a Static Random AccessMemory (SRAM) Distributed Arithmetic (DA) technique isused to synthesize precision baseband pulse-shapes [9, 10,11]. 40 and 20 percent square-root and full Nyquist raisedcosine waveforms have thus far been generated atbaseband by sampling and storing the appropriatecombinations of transmitted data patterns. The full Nyquist

pulse patterns were generated for testing purposes only. Thedata were sequentially generated for all symbol combinations

over specified symbol apertures. An aperture is the length oftime for a specified number of symbols to occur.

The frequency response of the square root raised cosinefunction (SRRCF) is defined by

H(f)

, 0 sl-I__(1 -p) + 4l]

_(1-I_) + 41_

where Ts is the symbol period in seconds, 13is the rotloff

rate, and fh is the half-amplitude frequency which equals1/2T s. Taking the inverse Fourier transform of equation H(f),

the impulse response is found to be

298

h(t) = ,, [,.,,.[.,,0,,]oo.[.,,.0,,1.[ 13] T, + T, JT_ !-. (4_t

_(1-_)+4 413t 1 t__s )2

where h(O)= 1 and

The rolloff rate B is expressed as a percentage and controlsthe trade-off between the bandwidth and pulse duration. Alower rolloff rate will allow for a narrower passband, but itsimpulse response will cause more ISI.

Aperture lengths of 6 and 12 symbols were investigateddepending on the modulation scheme involved. Data werecollected for 8, 16, and 32 samples per symbol. Eachsymbol was shaped with the SRRCF and the effects ofadjacent pulse responses were accounted for over thecenter symbol period. As the aperture length is an evennumber, the actual data collected represent the transitionfrom one symbol to the next. The amount of memory requiredto represent the full set of data on each orthogonal channelfor a given modulation format is

memory = sps x levels aperture

where sps is the number of samples per symbol and levels isthe number of amplitude levels on one axis of theconstellation. The amount of memory can be reduced byexploiting the symmetry in certain modulation formats. Theflexible aspect of this pulse shaping procedure is that datacan be generated for any rolloff rate, samples per symbol, orsymbol aperture length with high precision. A typicalsoftware generated eye-diagram is illustrated in Figure 7.

1.O

0,5

0.0

0°0 0.8 1.0 :L.E

Normalized Symbol Numb_

Figure 7: 8PSK Example Eye Diagram.

2.0

BASEBAND PRECOMPENSATIONEverything accomplished thus far assumes all subsequentsignal transformations are ideal linear operations. However,the nonlinear responses of devices such as TWTA's distortand degrade the signalling scheme. As a second orderapproximation, a narrow band, frequency independent,memoryless model of a TW'rA has Amplitude to Amplitude(AM/AM) and Amplitude to Phase (AM/PM) characteristicsthat can be modeled as

2r(t)A(r(t)) -

1 +r2(t)

• (r(t)) = D0 r2(t)l+r2(t)

where r(t) is the input amplitude level to the TWTA. Manytechniques such as precompensation, equalization, and ISIcancelling have been proposed to minimize this type ofnonlinearity [12, 13, 14]. In general, transmitterprecompensation can occur at baseband, IF or RF. IF andRF precompensation have been applied in many terrestrialmicrowave radio systems. The limitations observed in theseapplications are lack of flexibility, long term stability, anddifficulty of accurate construction. For these reasons, therehas been interest in applying digital signal processingtechniques to perform baseband precompensation.Amplitude and phase distortion for an operational TWTA canbe determined empirically and thus preciselyprecompensated. Most implementations thus far havefocussed on memoryless systems that warp the signallingconstellation with an inverse transform of the amplitude andphase response of the TWTA nonlinearity. However, insystems using pulse shaping, the r becomes r(t) causing thewarping to indtroduce additional ISI that results in degradedbit-error-rate performance and spectral regrowth. Thisproblem can be solved by using a memory basedarchitecture [15] similar to the pulse shaping technique. Thistechnique reduces ISI accurately to the symbol aperture. Itis interesting to note that this technique is well suited forsatellite communications that use lower order modulation

schemes. In terrestrial microwave systems, high ordermodulation schemes are typical. This puts a significantlimitation on the aperture and thus accuracy that can beobtained. The memory size as shown in the waveformsynthesis section depends exponentially on the number ofdiscrete amplitude levels of the constellation.

299

The precompensation isimplemented in terms of signalspace vectorsrelativeto each modulation scheme. Thevectorsarethe complex representationofthe in-phaseandquadrature sample values.The magnitude and phase ofthese vectors is precompensated for a set a data thatrepresents the AM/AM and AM/PM characteristics of thenonlinearity. The magnitude precompensation can beaccomplished by extending the linear portion of the AM/AMcurve. The input signal value enters the precompensator,and the output power is found relative to the linear extensionof the data. This value is then found on the actual datacurve, and the corrected input backoff corresponding to thisnew point is output to the rest of the system. Next, thecorrected input backofl is further corrected for the AM/PMcharacteristics of the curve. The phase shift correspondingto this input backoff is found, and this value is subtractedfrom the phase of the signal entering the TWTA. An exampleeye diagram of precompensated data for BPSK is shown inFigure 8. A square root 40 percent raised cosine pulse wasused to shape the data at 16 samples per symbol with atwleve symbol apeture. Measured AM/AM and AM/PM curvesfrom a TWTA were used to model the nonlinearity. Figure 10shows the precompensated scatter plot. As shown in Figure9 the precompensated data exhibits increased spectraloccupancy that must be included in the bandwidthcalculations for the modulation upconversion andamplification functions. The clustering and ISI effects seenin the eye diagram are products of the square root filter. Afull raised cosine or matched square root raised cosines donot exhibit these effects. Figure 11 shows the eye diagramafter the TWTA distortion and a matched receive finiteimpulse response filter with 96 taps.

This type of precompensation would be particularly useful inan operational system. A network using precompensationcould adapt and load new precompensation data as nonlinearchannel characteristics changed. Comparisons are beinggenerated that compare the total degradation for a set ofmodulation formats as a function of input backoff, filtering,and symbol aperture for baseband precompensated signals.

I .... I I .... IO.O O5 1,0 1,5 20

Normalized Symbol Numb_

Figure 8: Precompensated eye diagram.

• O! " ' '

Simul=ion Frequency (Hz)

Figure 9: Precompensated spectrum.

o.5

_o

b3

-05-

t t _ I ] I-_.5 -t.O -0.5 O,O O.5 1.O 1.5

Signa_ Real Pan

Figure10: Precompensated scatter plot.

0.6

04

0.2

0.0

-0,2

-0,4

-0.6

I .... | .... I .... I .... ]O,O O,$ t,O 1,6 2 • 0

Normalized Symbol Number

Figure 11: Receive filter output eye diagram.

FTMC TRANSMITTER HARDWAREPrototype hardware has been constructed to verify FTMCconcepts with real system nonidealities. The transmitterblock diagram is shown in Figure 12 and the FTMC modemchassis layout in Figure 13. Serial data can be clocked in atTTL levels continuously or in a bursted format. This data isthen translated to a parallel format of one to four bits wide.One bit is supplied to the rate 1/2 k=7 convolutional encoder,and the rest bypass it. Uncoded, coded, and burst controlbits (total of nine address bits) are supplied to a modulation

300

mapping function. This modulation mapper is a 512 X 4 bitSRAM loaded with four bits of information that represent fourin-phase and four quadrature levels. This information (eitherdata or preamble) is turned broadside to generate symbolapertures via shift registers. Symbol apertures aretransformed into pulse-shaped symbols with the Nyquistpattern generators. The pattern generators are each 128K X8 SRAM's. The SRAM's are loaded with the patternsgenerated via the techniques discussed in the waveformsynthesis section. The eight bit samples output from thefilter SRAMs are either directly output to the digital to analogconverters or used as address bit to the precompensator.The precompensator performs the nonlinear transformationas described in the precompensation section. The digital toanalog converters use eight bit quantization. Figures 14through 19 show eye diagrams and spectrums of four of themodulation formats, using full 40 percent raised cosinepulse shapes. In each figure, the sample rate is 32.768 MHzand there are 32 samples per symbol. This gives a bit rate of

Bit rate = sample rate Iog2(M))samples per symbol

Each spectrum is plotted in a frequency windowcorresponding to the width of the main lobe, 2/T b, of a BPSKsignal with the same bit rate, where Tb is the bit time.

!m

Burst Control I

Modul- _ Address

_I ]Decoder

Convolutional

Data In "_ S/P _ Encoder _ QLrl

r- r]LI

(Demodulator Lower)

:.=

.i.:.t

ADC (Upl_r ._>.

and Lower) "im,.

T:T

:i:i:i:................ ? ........................................ ........ i.......... i .......... T........ 11

Figure 13. FMC Chassis Mechanical Layout

17 ............

Analog

.......I ....... I I -- I 8_._and

_=----=-=_ Quadrature

Figure 12 : Digital Baseband Modulator/Encoder

The IF upconversion module translates the basebandmodulated signal to 70 MHz using standard quadratureupconversion techniques (Figure 20). The 70 MHz carrier isobtained from an analog signal generator that can be phaselocked to the symbol time to satisfy Q2pSK orthogonalityconditions. Direct Digital Synthesizers (DDS) could also beused to generate the IF carrier.

The quadrature spectral reconstruction filters are designedto accommodate the full range of FTMC operation in terms ofdata and sampling rate. The signal bandwidth (assuming nospectral shaping) is obtained as

BW = fssps

where fs is the sampling frequency and sps is the number ofsamples per symbol. The maximum bandwidth, BWmax is 5

MHz. This occurs when fs is 40 MHz and sps is 8 samples

per symbol. The spectral reconstruction filters must containthis frequency in their passband. The other constraintdefining the frequency response of the reconstruction filtersis the lowest alias frequency. This occurs at fs = 10 MHz andsps = 8, giving

falias =fs- fs = 10 MHz- 10 MHz = 8.75 MHzsps 8 sam/sym

This falias of 8.75 MHz must be in the stopband of the filterresponse. The resulting filter mask is shown in Figure 21,with the 48 dB of attenuation specification obtained from thequantization error of the 8 bit digital to analog converters.Note that the eventual limit of available sample rates anddata rates is limited by the lack of flexibility in thereconstruction filter. To maintain the low level of ISI obtained

by the Nyquist filters the group delay variation is limited to

301

Eye Diagram

gure16: 8PSK Eye Diagram

Figure18:16 QAM Eye Diagram

RF.J r -3,B 01_ ATTFN $iii dl

p_llt.11411 l_,d_ _I P''_rTIIr'rl|

' ,I 1

CENTER • H|

_4mlll IW _ leHZ

\

_AN 2.11411 14Hz

vl_ I kHl elWP l,l iIc

Figure 15: BPSKSpectrum

Fi i**ii i

CENT'ER II

Figure 17: 8PSK Spectrum

ii

SlIAN W.. t44 t4Hzviii | k),l| ||_11 |.iii io|

-I1.11 _ ATTIN $11 all

.. r,.1II, $1HI 14H=

, IL ,,11

C_TFJq • k_l p_'d III. Sgll l,t"_ZIW i kt-!_ VIIIW f. kh_ Olllm _I.III l.lC

Figure 19:16 QAM Spectrum

-_.t -r.B_el_nd Reconstruction ] '

In-Phase Filter

n

Baseband Reconstruclion ' _ 1

[ ' - Vector Modula or i

Variable Atlenuator

Tills( Point

Burst Control

Figure 20 • Analog IF Upconversion

302 C='_,G,NAL PAGE IS

C°F F'OOff QUALITY

15 nanoseconds up to 3.5 MHz. This means that 40 percentNyquist filters (cutoff approximately 3.5 MHz) will have agroup delay over their entire bandwidth of no more than

T,rnax- T-_ax6yrn_law_ 15 nS - 7.5 %

Tsyrn_ln_, 200 nS

for the worst case symbol rate of 5 Msps. This effect causesan expected Eb/N o degradation of no more than a few tenthsof a dB.

reference

level

-48 dB

_ variation : 15 nS from 0 to 3.5 MHz

0 1 2 3 4 5 6 7 B 9 t0

MHz

Figure 21 : Spectral Reconstruction Filter Mask

CONCLUSIONThe architecture and hardware prototype of a flexible trellis-coded modem and codec transmitter has been presented.The FTMC transmitter uses digital modulation and coding,multi-symbol digital pulse shaping, and digitalprecompensation to tradeoff complexity for improved signalquality, bandwidth, and power efficiency. This type ofarchitecture is envisioned to be of use in FDMA satelliteuplinks and TDM or CDM satellite downlinks.

tn any ptanned operational satettite network, the choice ofmodulation and coding techniques is dictated heavily bynumerous other system concerns. However, offering futuresystem designers workable flexible modems and codecsmay relieve specifications in other subsystems and offer anenhanced satellite system that can offer more varied andhigher quality services.

ACKNOWLEDGEMENTThe authors would like to thank L. Shimoda of Ohio

University for generating the simulation data for Figures 8through 11.

REFERENCES[1] Viterbi, A. J., et al: A Pragmatic Approach to Trellis-Coded Modulation. IEEE Communications Magazine, July1989, pp. 11-19.

[2] Harrold, J. L.; Budinger, J. M.; and Stevens,G. H.: On-Board Switching and Processing, Proc. IEEE, vol. 78, No. 7,July 1990, pp.1206-1212.

[3] Nuspl, P. P., et ah On-board Processing forCommunications Satellites : Systems and Benefits,International Journal of Satellite Communications, Vol. 5,1987, pp. 65-76.

[4] Campanella, S. J.: Communications Satellites: Orbitinginto the 90's, IEEE Spectrum Magazine, August 1990, pp.49-52.

[5] Hansen, J. W., et el: US TWTs from 1 to 100 GHz,Microwave Journal • 1989 State of the Art Reference, pp.179- 193.

[6] Saha, D.; and Birdsall, T. G.: Quadrature-QuadraturePhase-Shift Keying, IEEE Transactions on Communications,Vol. 37, No. 5, May 1990, pp. 437- 448.

[7] Sklar, B.: Digital Communications Fundamentals andApptications, Prentice Hal/, 1988.

[8] Lee, P. J.: Computation of the Bit Error Rate of CoherentM-ary PSK with Gray Code Bit Mapping, IEEE Transactionson Communication, Vol. 34, No. 5, May 1986, pp 488- 491.

[9] Poklemba, J. J: Programmable Digital Modem, NASASpace Communications Technology Conference, November1991.

[10] Aghvami, A. H.; and Gemikonakli, O.: 16-ary QAMSystem for Low Data-Rate Satellite Services, Electron. Lett.,Vol. 25, No. 16, August 3, 1989, pp. 1055- 1056.

[11] Allan, R. D.; Bramwell, J. R.; and Tomlinson, M.: A HighPerformance Satellite Data Modem Using Real-Time DigitalSignal Processing. Olympus Utilisation Conference, Vienna,Italy, May 1989, pp. 51-55.

[12] Karam, G.; and Sari, H.: Analysis of Predistortion,Equalization, and ISI Cancellation Techniques in DigitalRadio Systems with Nonlinear Transmit Amplifiers.

[13] Biglieri, E.; Barberis, S.; and Catena, M.: Analysis andCompensation of Nonlinearities in Digital TransmissionSystems, IEEE Journal on Selected Areas inCommunications, Vol. 6, No. 1, January 1988, pp. 42- 51.

[14] Cheung, S. W.; Aghvami, A.H.: Performance of a 16-aryDEQAM Modem Employing a baseband or RF PredistorterOver a Regenerative Satellite Link, lEE Proceedings, Vol.135, No. 6, December 1988, pp. 547-557.

[15] Karam, G.; and Sari, H.: A Data Predistortion Techniquewith Memory for QAM Radio Systems, IEEE Transactions onCommunications, Vol. 39 , No. 2, February 1991, pp. 336-344.

3O3

Form ApprovedREPORT DOCUMENTATION PAGE OMBNoOZO*Ol_

PublicropodJngb_ tot_ie oo_lon ol L,ttocmmtionhioMimslo_to avtrllge1 hourperro_, btdudlirl_I_ _ torreYiewtng_, eearohblgoxi|_ngdll_ _s,oathecingand mai_ainm,g Itw data_. and¢_'n_ng _ reviewing_ o_lec_o_of Infom-,4tmn.,Send_ regardingINs burden,Mlima_ or Imy _ _q>ectolhm¢oikmbon¢4kltownetlcm,includingsuggela_n=for redu¢/_ INs buKlen,to Wuhtngto_HeadquartersS4m_ou.DirectorateforIrdormaPonOpem_ anti Reports,1215JefforlonDe'_,sHighwly,Suite1204. Arlington.VA 22202-4302.m'_dto _ Officeo¢Mant_ementrand_t, P_eq_,_rtt_ Project(0704-O188),Washington,DC 20503.

1. AGENCY USE ONLY (Leave blank} 2. REPORT DATE 3. REPORT TYPE AND DATE=; COVERED

November 1991 Conference Publication

4. TITLE AND SUBTITLE 5. FUNDING NUMBERS

Space Commumcatio_s Technology Conference

Onboard Processing and Switching

6. AUTHOR(S)

7. PERFORMING ORGANIZATION NAME(=;) AND ADDRE=;S(ES)

National Aeronautics and Space Administration

Lewis Research Center

Cleveland, Ohio 44135- 3191

9. SPONSORING/MONITORING AGENCY NAMES(S) AND ADDRESS(E=;)

National Aeronautics and Space Administration

Washington, D.C. 20546-0001

WU-650-60- 21

8. PERFORMING ORGANIZATIONREPORT NUMBER

E-6548

10. SPONSORING/MONITORINGAGENCY REPORTNUMBER

NASA CP-3132

11. SUPPLEMENTARY NOTE=;

Responsible pe_on, James M. Budmger,(216)433-3496.

12a. DI=;TRIBUTION/AVAILABILITY STATEMENT

Unclassified - Unlimited

Subject Categories 17 and 32

12b. DISTRIBUTION CODE

13. ABSTRACT (Maximum 200 wo_a)

This proceedings of the NASA Space Communications Technology Conference held oil November 12-14, 1991, in Cleveland, Ohio, _ts

onboard processing mad switching t_chnologies for applicalion in fumt_ satellite communicatioos systexns. 0nboard processing for B-ISDN

(broadband integrated services digital network) services, packet-switched P"DM,VIDM (frequency division multiple access/time division), a

novel variation of CDMA (cod_ division multiple aCCeSS),and the merits of OBP (onboard processing) for low Earth orbiting mobile systems are

addressed in the section, Sate/lit_ Network Architectures. Telecommunications services for the next generation of networks, a traffic analysis ofB-ISDN, Klva.nced lighiwlveumtworks,and s_rvivab[esystemp¢ocessi_tgarecovex_ inNetwork Controland ProtocoLs.Papexson bendwidth

efficient coding, artificial intelligence for Earth tetminels, multicarde, r demodulailon, neund-netwofk-based video compression, an expert system

for fault diagnosis, and laboratory measurements of INTELSAT OBP subsystems are ix_mted in Concurreat Poster P_esentations end Demon-

stratioos. Two papers on expert systems for space hardware diagnostics and fault tolerance applied to fiber-optic networks and multichannel

demulilplexea's are covered in Fault Tolerance and Autonomy. Digital, optical, and acousto-.optical approaches to multichannel demultiplexing

and demodulationfor onlx_-,d processed FDMA are presented in Multichannel DeJnultiplexing and Demodudadon. In Information Switching andRouting, an onboard baseband switch, an OBP payload for asynchronous relay satellite networks, OBP for future telecommunications services,

and congestion conurol for satellite packet switching are presented. In Modulation and Coding, _ final section of this document, advanced

modulation and coding technologies are applied to a B-ISDN modem-code,:, a flexible, high-speed codec, and two variable-rate digital modems.Decoding multilevel demodulation and two approaches to digital pulse-shaped modulation for nonlinear satellite communications are also

presented. Papers for the Final two sessions of the conference were not available at Ihe time of publication. These sessions, which addressed

planned communications satellite systems under development at INTELSAT, NASA, Motorola, and Space Syst_ms/Loral, and a panel session on

the issues and challenges of onboard processing and switching technology insertion, are covered in a companion publication.

'14. SUBJECT TERMS

Onboatd processing; Modulation; Coding; Autonomy; Switching;

Routing; Network control

17. SECURITY CLASSIFICATION 10. SECURITY CLASSIFICATIONOF REPORT OF THIS PAGE

Unclassified Unclassified

NSN 7540-01-280-5500

IS. NUMBER OF PAGES312

15. PRICE CODE

A14

19. SEC_JRITYCLASSIFICATION 20. LIMITATION OF ABSTRACTOF ABSTRACT

Unclassified

Standard Form 298 (Rev. 2-89)Prescribed by ANSI Std. Z39-18298-102

NASA-Lamgley,1991