Sie sind auf Seite 1von 8

. .

e -

. ,

SPREAD SmPECTRUM COMMUNICATIONSMYTHS AND REALITIES


Andrew J. Viterbi

Coding is always beneficial and sometimes crucial for the suppression of interference in spread spectrum communications.
INTRODUCTION PURPOSES
The purpose and applicability of spread spectrum techSpread spectrum communication techniques date niques is threefold: back to the early fifties. Since the earliest applications, Interference Suppression systemimprovementshavebeenmoreevolutionary iEnergy Density Reduction thanrevolutionary. Like most improvements in elecRanging or Time Delay Measurement tronic systems, these are due primarily to the availability Foremost among theseis the suppression of interferof ever higher speedintegrated circuit components, of ence.which may be characterized as any combination which translate in this case to wider spread spectra. In the following: three decades the achievable spreading factor has grown 1) Other Users: intentional (hostile or unintentional), by about three orders of magnitude to the point thatwe 2) Multiple Access:spectrumsharing by coordiare now limited more by bandwidthallocationsthan by technology limitations. Before we examine the quanti- nated users, 3) Multipath: self-jamming by delayed signal. tative effectsof spreading, let us catalog briefly the multiProtection against in-band interference is usually ple purposes of spread spectrum communications. called anti-jamming (A/J). This is the single most extento expansion First, we note that spreading here refers sive application o f spread spectrum communication. A of the bandwidth well beyond what is required to transis that o f multiple access by numerous similar application mit digital data. Thus, a system transmitting data at a rate users who share the same spectrum in a coordinated ( R )of 100 Mbits/s using approximately 100 MHz o f bandmanner, in that each employs signaling characteristics or width (W) is not spread at all, whilea system transmitting parameters (often referred to as codes) which are distinat 100 bits/s spread over a spectrumof about 100 MHz of all other users. One reasonfor guishable from those has a factor W / R = 106, or 60 dB o f so-calledprocessing using this shared spectrum, so-called code-division multigain. ple access (CDMA), is that by distinguishing signals in this way, separation in the more common dimensions of frequency or time is not required, and hence the usual The authoris with the Linkabit Corporation, SanDiego, CA92121. transmission tolerances need not be imposed on these Which parallels the evolution of data rate capabilities of digital parameters. communications.

May 1979

11

The third formof interference suppressed by spread very reliable. Rather, the spread spectrum signal used for spectrum techniques is the self interference caused by ranging is a long sequence o f polarity changes (binary ofthe signal, arriving multipath in which delayed versions PSK-modulated signal). Upon reception, this is correvia alternate paths, interfere with the direct path translated against a local replica and lined up to perform an mission. accurate range ordelay measurement. While thesecondandthirdforms of interference BASIC TECHNIQUES would appear morebenign than thato f a hostile emitter, the technique and effect are the same. What makes the Havingoutlined themultiple uses o f spectrum spreadintentional interference more challenging is the game ing, we must examine at least a superficial description of aspect o f the problem and the fact that the interfering the concept beforewe can proceed to dispel myths and source is generally granted much more power than the uncover realities about this increasingly popular techcommunicator, which is usually not the case nique. Fig. 2 is an all-purpose diagram for cooperto describe spread ating users and even less so for multipath interference. spectrum modulation. Multiplication of two unrelated The second class o f applications centers about the is the convolusignals produces signal a whose spectrum reduction o f the energy density o f the transmitted signal. tion o f the spectra o f the two component signals. Thus, if This, too, has a threefold purpose: the digital data (binary)signal is relatively narrow-band 1) to meet international allocations regulations, compared to the spreading signal, the product signal will 2) to minimize detectability, have nearly the spectrum o f the wider (spreading) signal. 3) for privacy. So much for the modulator. At the demodulator, the Downlink transmissionsfromsatellitesmustmeet received signal is multiplied by exactly the same spreadinternational regulations on the spectral density o f the ing signal. Now if the spreadingsignal, locally generated signals received on earth. By spreading this energy over at thereceiver, is lined up(synchronized) with the a wider bandwidth,totaltransmittedpowercanbe received spread signal, the result is the original signal increased, and hence performance improved. Spreading plus, possibly, some spurious higher frequency compoalso decreases the detectability of a signal by a regulatory nents outside the bandof the original signal, and hence body whichemploys spectral analysis to monitor or regu- easily filtered to reproduce the original data essentially late emissions. (It is not known whether bootleg radio undistorted. If there is anyundesired signal atthe amateursare usingspreadspectrummodulation to receiver, on the other hand, the spreading signal will affect it just as it did the original signal at the transmitter. evade FCC regulations.) Even more promising is the Thus, even if it is a narrow-band signal in the middle o potentialforachievingprivacy in communication by f spreading one.s signal sufficiently to hide in the backthe bandof interest, it will be spread to the bandwidth o f ground noise. the spreading signal. The application of spread spectrum for ranging or The result is that the undesired (jamming)signal will position location is rapidly gaining in importance. In simhave a bandwidtho f at least W . If its power is J watts, its average density, which is essentially uniform and can be f measuring the plest terms, position location consists o treated as wide-band noise,will be delay o f a pulse or pulses. Error in delay measurement is inversely proportional to the bandwidth of the signal NO= J/W watts/Hz. f pulse. This is most easily seen by the simple exampleo Let the desired component o f the received signal have Fig. 1.The accuracy o f the measurement At is obviously powers watts. Thus,if the data rate isR bits/second, the proportional to therise time o f the pulse, which is inverreceived energy per bit is o f the pulse signal. Of sely proportionalto the bandwidth course, a one-shot measurement ona single pulse is not Eb = S / R watts . second. . - .. . - _-.
...~...I

. I

Now it is generally recognized that digital communication system bit error rate performance is a direct function of the dimensionless ratio Eb/No, which for spread spectrum signals may thus be expressed as Eb No

-s w
J R

and hence, the jamming power-to-signal power ratio is

! . .

Uncertainty At = Rtse Time of Pulse = ...:. ~ . .

.Wl

Fig. 1. Time delay measurement.

This establishes that ifEb/No is the minimum bit energyto-noise density ratio needed to support given a bit error rate, and if W / R is the ratio o f spread bandwidth to the IEEE Communications Magazine

12

Signal Data
I

Rate R
U

Spreading Signal
I

Spreading Signal

!
I

Multiplication ONCE Spreads Signal Bandwidth Multiplication TWICE Followedby Filtering Recovers Original Signal DESIRED Signal Multiplied TWICE, Undesired ONCE . . , . ..,______. __ . . ___. . _ _ . Fig. 2. Basic spread-spectrum techniques.

.-

_---____.,_.I

..

_ I -

original data bandwidth, also called the processing gain, then J/S is the maximum tolerable jamming power-tosignal power ratio, also known as the jammingmargin. We have come this far without even specifying the f the spreading signal. There are, in fact, characteristics o two distinct classes o f spreading techniques. The first is calleddirect sequence or pseudonoise (PN)spread specis achieved by multiplication by trum. Here the spreading a binary pseudorandom sequence whose symbol (switching) rate is many times the binary data bit rate. The is sometimes called the spreading sequence symbol rate chip rate. The second class utilizes afrequency hopping carrier. Here the spreading signal remains at a given frequency for each bit or evenfor several bits. Thus, locally it is no wider than the data signal, but when it hops to a new frequency, it may be anywhere within the spreading bandwidth W. One fundamental difference between the two techniques is that direct sequence PN spread signals can be coherently demodulated. With frequency hopped signals, on the other hand, phase coherence is difficult to maintain when the signal frequency is hopped over a wide range; hence, this modulation is usually demodulated noncoherently. We are now ready to explore severalfirmly entrenched items of common wisdom regardingthe relative desirability of variousfeatures of spread spectrum systems. Often these attitudes are misguided, a s we shall presently show. In all cases, the ideas hold for both classes of spread spectrum techniques, but for all but the last concept the arguments are somewhat simpler for direct sequence spreading, which we shall therefore consider. We are ready now to reveal the first of four myths. First Myth: Error-correcting coding requires redundancy, wnich spreads bandwidth and thus reduces auailable processing gain for the available bandwidth.

Reality is, in fact,just the opposite. To see that coding does not reduce the effective processinggain, let us f the symbof rate rewrite jamming margin (1)in terms o R, and thesymbol energy E,. These are related to the bit rate and thebit energy through the code rate r, defined as the number of data bits per transmitted symbol, or the inverse of the coding expansion factor. (For example, a rate 1/2 coded system transmits two code symbols for each data bit.)It follows that R, =R/r and

E,

= Ebr.

Now if we repeat the previous dimensional argument replacing bits by symbols everywhere, we have

J/S

=-

WIR,

,/No

but substituting the preceding definitions for symbol rate and energy, we obtain for the maximum tolerable J/S ratio W No W/R J/S = --R/r Ebr &/No
f which gets us back to (1).This may seem like sleight o hand, but it really is not. Moreover, although it will take some further reading to be convinced, we are really ahead o f the game. For with coding, the required &/NO for agiven level o f performance (bit error rate) is actually reduced. Thus, for a given processing gain (W/R) the jamming margin is further increased by coding. For those who aresatisfied that spectrum spreading (especially direct sequence PN) techniques make the noise look white while the signal energy, without or with coding, can be fully recovered by the receivers correlating multiplier, the dispelling o f the First Myth will come as no surprise. Yet it is often this sophisticated group who will fall prey to the

Second Myth: Error-correcting coding is effectiue only against uniform interference.


2Symbol rate refers tothecode symbol of theerror-correcting f the PN spreading code, which is usually called chip code-not that o rate.

Coding does not reduce theeffective processing gain in a spread spectrumsystem.

May 1979

13

In particular, the myth continues; coding is not effective against pulsed interference. Yet, this is even more dramatically false thantheFirstMyth.Letus consider what the effect of pulsed interference can be for an uncoded system. Suppose the jamming is present only a fraction p < 1 o f the time, but that during this time, the noise density level is increased to a level No/p watts. This assumes spectrum spreading which turns the jamming signal into broad-band noise and an average powerratherthan a peakpower limitation onthe jammer. (While this may be slightly pessimistic for the communicator,anyotherassumption is a risky bet against technological progress.) Now it is well known that with coherent demodulation an uncoded BPSK modulated system produces a bit error rate P b related to &/No as

Clearly, the jammer would choose the duty factor p which pessimizes performance-that is, maximizes bit error rate. In terms of the approximation,which is a strict upper bound,-this occurs when 1
P=-

&/No

provided !,/NO

>I

at which value

where

Q(x) = -

5 6
1

"

e-""* du.

But if the noise is intermittent, and hence onlywith probability p corrupts a given transmitted bit3 with the higher noise density No/p, the resulting bit error rate becomes
3We assume for simplicity that a given interference pulse corrupts an, integral number of bits. This is a reasonable assumption if the pulse width is many times the bit duration. Otherwise,the situation is actually less favorable to the jammer.

(Note that although we worked with the approximation forits simplicity, hadweusedtheexact(Q-errorfunction) expression, the worst casep would be nearly the same and the maximum bit error rate would not be significantly lower.) The result is quite .dramatic. Pulse jamming-with spread spectrum but without coding-changes an exponential relation into an inverse-linear one. Numerically, if we desire bit error rate performance on the order o f P b io-', stationarynoise(orjamming)requires only &,/No = 10 dB, while with pulse jammingwe must have f Eb/No = 45 dB, an increase in required signal powero over three orders of magnitude! Amazingly, coding can almost fully restore this deplorable situation, but before we can explain why, we must briefly explore a summary of the general capabilities of coded systems.
~~ ~

Coding Fundamentals
When a binary data streammust be transmitted over anoisy channel with a troublesome bit error rate,coding can beused to significantly reduce the error rateincurred by the message. message bit stream is partitioned into blocks o f k bits, where kis the block length.Each such message block In block codingschemes, the is replaced with ann bit code word (n is bigger thank) which is transmitted in its place. Thus, every n bit transmitted"contains" only k message f the code is k/n bits per code symbol. bits so that the rate r o A common noisy channel model is the so-called Gaussian channel where each bit, viewed as a square pulse of amplitude k 1, is independently subjected to additive noise and an error occurs when the noise alters the pulse polarity. f errors, thereceived n-bit block can be any o f 2" possible words. Sincethere areonly Zk different code words that could have As a result o been transmitted (one for each k-bit gnessage block) and Zk is typically much less than 2". the number of possible received words 2" is much o f code words 2 k . For each received code word, the decoderdecides what was the most likely code word that was greater than the number transmitted, and the receiver then identifies the corresponding k-bit message block. In this way error correction can be achieved. Here theincoming message bit stream is applied to aK-stage shift register which is A notablydifferent approach is convolutional coding. shifted b bits at a time. For eachK message bits stored in the register, there arenlinear logic circuits which operate on the register contents to f the encoded output stream. For each shift o f the register, bnew message bits are inserted and n code bits are delivered, produce n code bits o so that the rateis b/n information bits per code symbol. In this case, a particular code bit depends on K message bits where K is called the constraint lengthof the code.Note also that a particular messagebit remains in the register for K/b shifts, and thusinfluences the value of nK/b code bits. f the entire Unlike block coding, the optimal decoding operation for convolutional codes requires a memorythat stores, in effect, a function o past history of the received bit stream. The performance (as measured by error rate) of a convolutional coding system improves as the complexity (Le., memory) allowed for the decoder is increased. Several methods of decoding convolutional codes have been developed. The optimal (maximumlikelihood) scheme is generally known as the Viterbi algorithm. Viterbi decoding for reasonably short constraintlengths is feasible to implement and high decoding speeds areachievable. For extremely low error probabilities, a large constraint IengthK is required. The computational complexity of Viterbi decoding for large K makes this approach impractical. Another approach, sequential decoding, then becomes more attractive. A thirdtechnique, feedback decoding, though inferior in performance against random errors, is particularly well suited to correcting systematic error bursts which may occur in fading channels. Both fading and pulse jamming introduce memoryin the channel and further modify the channel statistics.Yet the same coding techniques as used for the Gaussian channel are at least as effective here, provided interleaving is employed to reduce or eliminate this memory.

14

Magazine

IEEE Communications

ERROR CORRECTING CODING FUNDAMENTALS All that we need to know about coding for the present
purpose is that for practically any memoryless channel, f rate r bitslsymbol there are many good binary codes o for which the bit error rate is upper bounded [1 1by either
P b
U
0
1

< 2-K(a-),
2-Ka

E!
I

if it is a block code o f block length K

(4)

or
p b

< [1 - 2-a-l12

if it is aconvolutional code o f constraint length K.

(5)
_=_

In either case
ro r provided the code rater < ro < 1. Performance then depends strongly on the value o f the parameter4ro. This parameter, and consequently a, is increased if the decoder is furnished with everything the receiver knows about the channel; that is, for binary symbols, not only the receivers belief that the transmitted symbol was a zero or a one,,but how strongly the receiver believes this. This confidence in the decision is called asoft decision,quality information, or simply a metric. Now for a uniform Gaussian channel, with soft decisions furnished to the decoder, the all-important parameter ro is a function of only the symbol energy-to-noise density. That is,
CY=->l
ro = 1 - log, [ 1

ro

ar

Fig. 3. */No requirement in additive uniform noise.

(4) and (5) areonly upper bounds, and hence pessimis-

f coding, they are sufficient to tic estimates of the value o establish its merit, and in fact are reasonably accurate. As a practical matter, commonly used commercially and available convolutional decoders require&/NO an which is about 0.5-2 dB above the curve of Fig. 3 to achieve
P b
2 :

+ e-E~N~l]

(7)
(8)

where, as previously defined,

E, = Ebr.
ro as

Combining(6), (71, and (8),we can relateEb/No t o a and

For specific comparison, wehavefrom(2)that to achieve P b = on the uniform Gaussian channel without coding requires &,/NO = 9.6 dB. With rate 1/2 convolutionalcodes,practicalsoft-decisiondecoders require between 3.5 and 4.5 dB for the same performance. Rate 3/4 codes require approximately 1dB more. If the decoderis provided with only hard-decision inputs, it requires approximately 2 dB higher Eb/NO. Thus farwehave considered only stationary wideband noise, whether o f thermal origin, or so rendered by the direct sequence spread spectrum technique we have investigated. We now return to the pulse jammer and show howcodingremediesthedeplorablesituation which we left before the present digression.

No

_C Y
E b

In (21-nl
ro

1)

(9)

SECOND MYTH REVISITEDAPPROPRIATELY ARMED


Suppose now that we code as before,but nonuniform (pulse) jammer. Spreading causes for the this to

This quantity expressed in decibels (10 log &,/No - 10 log a)is plotted in Fig. 3 as a function of ro.Of course, Fig. 3 or (9) is meaningful only wheh taken together with (4) or (5). The interpretation is that for a given acceptable complexity o f implementation, which is roughly proportional to 2K for either class o f codes, and a given code rate r, we need to select a value of a to guarantee the required Pb, according to (4) or (5). This establishes ro according to (6), and finally we obtainEb/No by adding 10 log a to the ordinate of Fig. 3 for the given ro. If the resulting r o is greater than 1, we must choose a smaller code rater for which ro < 1 for the requireda.Although

May 1979

15

SCRAMBLER CODER INTERLEAVER


I

OR OR DEINTERLEAVER
I

>
....

UNSCRAMBLER DECODER
~~

j 1 I

i
!
i

L-.

. -

_____....___.^_.._

. . _._._.__.,._I____^.-I .

Fig. 4. Introduction of interleaving for disposing bursts.

appear at the receiver as wide-band noise o f density level N o / p , but for areduceddutyfactor p . Suppose, as before, with little loss of reality, that anintegral number o f code symbols are affected by jamming. We cannot quite apply what we just learned about coding because the jamming pulsesaffect many contiguous symbols;so we cpn hardlycall the channel memorylessas required. But this 'is easily remedied. Suppose we construct a device which randomlyscramblestheorder of the symbols priorto transmission, but after coding, and puts them back in the right order after reception, but before decoding (Fig. 4). (Scramblers and unscramblers are more commonlycalled interleavers and deinterleavers.) Buttheunscrambler which restoresthetransmitted symbols to their right place in order actually scrambles the regular jamming pulses into random patterns? Scrambling or interleaving thus makes our system memorylessagainand we canapplyournew-found coding knowledge. Withoutbelaboringtheexactdetails,arguingintuitively and believingly on the basis of (2) and (2'), let US replace e-E,/Nn by pe-pE5'N'' in all theformulas of the previous (uniform noise) section. Thus, (7) is replaced by ro = 1 - 10% [I

(9).This says that ifro is small enough, n o p h l t y is paid to a pulse jammer. Even for ro > 0.548, the penalty is small, as seen in Fig. 5, which shows the new pulsed noise case and reproduces the uniform noise case from Fig. 3. Thus, even more amazing than the original 35 dB loss to pulsed noise, coding recovers it all and then some. In mite o f this. there still are skeDtics who believe in the Third Myth: Interleaving destroys memory. Memory can be exploited to correct errors. Hence, interleaving is bad. The discussion leading to the system design of Fig. 3 should suffice to dispel such misgivings,particularly when we recall that soft decisions contain about all the information available aboutthechannel(granted, o f course, that memory may be exploited to extract such quality information). To definitely put to rest all discussion of the matter, let us consider a burst interference phenomenon which always affects B symbols, and let there be no overlaps o f bursts (overlaps can only helpthe communicator). In this case, the key parameter of (7') becomes

+ pe-pE,'No].

(7')

Combining (7') with ( 6 )and (8), we get

which causes ( 9 )to be replaced by

(9")

whichobviously reduces to (9) for uniformjamming ( P = 1). This is maximized by a jammer with duty cycle = (21-r" 1)e (10) provided ro > 1 - log, (1 Eb ae-' -max-= O+<I NO r0(2'-"' - 1)

which leads to the plots of Fig. 6 . Clearly (7") and (9") reduce to (7') and (9') forB = 1,which is the case when interleaving is employed. is even worse than shown in Fig. 6 . In fact, the situation

+ e-')= 0.548 for which


for ro

> 0.548.

Even the severe impact of pulsed jamming can be entirely contained by suitable coding.

If ro G 0.548, p = 1 maximizesEb/No and (9') reduces to


5Note the parallelwith spectrum spreading; the second multiplier unspread the signaland spread the interference; here the second device unscrambles the desired code sequence and scrambles the undesired pulsed interference.

-=-

r,

ar

Fig. 5. &/No requirement in pulsed and uniform noise.

16

IEEE Communications Magazine

p b

(9 < [1 -4 1
W K O

convolutional codes

(5')

z
I

which reduce to (4) and (5)whenq = 2. There is nothing to gain by using multiple signal alphabets for coherent, direct sequence systems, but with frequency-hopped systems, we can show[2] that for the worst case partialband jammer

cn
J

-I_- ro

1
ar

Fig. 6. */No requirement in pulsed noise without interleaving.

For as r decreases, the numbero f symbols per bit and, for a constant duration interference pulse, the number of symbols per pulse(B) increases. Thus, keepingB fixed, as is done for convenience in Fig. 6 , gives misleadingly favorable results. Our final "myth" happens fortuitously to coincide with reality. We call it, therefore, a "Folk Theorem." It concerns an interesting comparison of direct sequence (coherent) spreading with frequency hopping (noncoherent) spreading: Fourth Folk Theorem: (Myth = Reality) Performance of frequency-hoppedspreadspectrum is 3 dB worse than that of direct sequence (PN) spread spectrum. The commonly invoked "mythical" argument is that noncoherentsystemscan utilize atbestorthogonal signals (e.g., binary FSK modulation) insteado f antipodal signals (binary PSK modulation) and this accounts for the 3 dB. The trouble with this argument is that it ignores the possiblity o f higher signaling alphabets(such as MFSK) and, worse still, the real possibility that frequencyhopped systemsmay be morevulnerable to nonuniform interference. We note, in fact, that in frequency-hopped systems, the jammer need not pay the cost o f a higher peak power signal, for if he jams just a fraction o f the band,6p < 1, with power density N o / p , he will appear to thereceiver just as a partial time jammer. Note also that if the hopping rate is at least as great as the symbol rate, interleaving is unnecessary. o f sizeq, (4) and (5)must be modified to For alphabets become p b < q-K(h--') block codes (4') and
6Possibly varyingthisbyhoppinghimself in order to defeat the obvious communicator strategy of determining the jammed region and staying out of it.

provided

For asymptotically large q , this approaches


Eb _ - (4 In 2)a

No

which is exactly a factor o f 2 (3 dB) above the minimum of Fig. 3 which occurs as .ro -0. This comparison is shown in Fig. 7, which also shows the diminishing returns of using alphabet size q > 8. The asymptotic minimum is virtually reached for q > 32. Notice that noncoherent frequency-hopped systems exhibit a minimum, while coherent direct sequence systems improve monotonically as l/ro increases. The explanation o f this behavior is better understood by examining Fig. 8 which is a more detailed and more realistic examination of performance for an octal alphabet. Here the assumptions are more' realistic. Specifically, channel quality information is limited to two bits (four levels) out o f each of the q matched filters. This also allows for apracticalautomatic gain control (AGC) technique, which has not been mentioned up to this point. The curve for p = 1 is, o f course, for uniform noise. The increase athigh rates (r andro close to 1) is due to the lack o f coding redundancy. The increase low atrates is due to the higher loss, characteristic o f noncoherent combining o f symbols in high diversity (here low rate) noncoherent communication systems.As p , the fraction of the interference bandwidth decreases, performance diversity is gets increasingly worse at high rates since the lacking toovercomethestrongjammer. But as ro decreases, diversity becomes sufficient to fully defeat the low p jammer, as shown by the family of curves o f Fig. 8.

May 1979

17

spreading of one or more orders of magnitude over direct sequence spreading technology which greatly overshadows the system edgeof the latter.

Although the efficiency of direct sequence spread spectrum is about double that of frequency hopping, this advantage is overshadowed by the greater band spreading achievable with frequency hopping technology.

REFERENCES
ii.-1

.I

. .

1:3

.-

I-.

! 1 .
5

.
I._

i i _i.~Il_i 7 . 8 9

_=_

[ 11 A. J. Viterbi and J. K. Omura, Principles of Digital Communication and Coding. New York: McGraw-Hill, 1979. [2] A. J. Viterbi and I. M. Jacobs, Advances in coding and modulation for noncoherent channelsaffected by fading, partial-band and multipleaccess interference, in Advances in Communication Systems, VOI. 4. New York: Academic. 1975

ro

ar

Fig. 7.

requirement for noncoherent frequency hopped systems.

SUMMARY
Beyond cataloging the many uses of spread spectrum communication, we have made no attempt to uniform be field. in our treatmentof this extensive and many-faceted We have concentrated rather on its application for the suppression of interference, and have made three main points: Coding is alwaysuseful, and it may be critical to adequateperformance o f spreadspectrumsystems, is partialparticularly when the natureof the interference f frequency-hoppedspreading, time,or in the case o partial band. Proper interfacing of the decoder to the demodulator, in utilizing quality (soft decision) information, is i m p o r t a n t to ensure m a x i m u m benefit f r o m coding. Interleaving or scrambling may be equally essential in the presence of burst interference. Direct sequence spread spectrum efficiency is about double thatfor frequency hopping. This is tantamount to doubling the processing gain W/R. However, frequencyhopping technologymay have an edge in achievable band

Whats Your Reaction?

Now that you heard what Andrew Viterbi has to say, what do you have to say? Praise, criticism, corrections, disagreements, and other comments are always welcome and appreciated. (Please indicate if we may publish your remarks in the Packets to the Editor column.)

-Editor

L1

__ L I
2

I^___;
ru

i . -

-.

Jt
4

1 i.

.A i .-

3 1 1 - _ar

L ,L .. ,...J 6 7 8 9 1 0

Fig. 8. &/No requirement for octalnoncoherentfrequency hopped system in partial band interference with receiver quantization ( p = interference fractional bandwidth).

18

IEEE Communications Magazine

Das könnte Ihnen auch gefallen