Vous êtes sur la page 1sur 8

SYSTEM DESIGN CONSIDERATIONS FOR THE USE OF TURBO CODES IN AERONAUTICAL SATELLITE COMMUNCATIONS

E.F. C. LaBerge, Honeywell


key to this near-optimum performance lies in the iterated decoding of the received data stream and in the parallel concatenation of two relatively simple systematic encoders separated by an interleaver. Since their first description, Turbo Codes have been a hot item for coding researchers, and the available technical material has exploded in the last few years. A detailed description of the "hows" and "whys" of Turbo Code performance is not possible in this paper, but those who are interested can find a wealth of tutorial information in [2]-[4]. We do require a brief introduction to Turbo Code operation and performance, however, to allow us to discuss their application to aeronautical SATCOM communications.

Introduction
So-called "Turbo-Codes'' currently represent one of the most active areas of research in Coding Theory. Initially described by Berrou, Gavieux, and Thitimajshima in the early 1990s, Turbo Codes have performance that approaches the Shannon limit. The key to this near-optimum performance lies in the iterated decoding of the received data stream. Turbo Codes are a superset of block and convolutional codes familiar to most digital communications engineers. Despite their obvious advantages, however, Turbo Codes do have some limitations that make them more or less suited to specific applications. This paper examines the utility of Turbo Codes for aeronautical satellite communications. A brief introduction to Turbo Code principals is followed by a high level examination of Turbo Code properties, and presentation of simple examples of Turbo Code applications to aeronautical communications.

Parallel Concatenated Codes


The Turbo Code encoder (Figure 1) consists of the parallel concatenation of multiple simple encoders, separated by an interleaver that reorders the input data. Incoming data is partitioned into blocks of length N corresponding to the depth of the interleaver. Because the encoders are systematic the data stream is present without modification in the encoder output, and need only be repeated a single time. The N-Point interleavers ensure that the probability that a different bit sequence is presented to each encoder is quite high. Thus, the
Output Data Bits @ : v

n r b o Codes
In 1948, Claude Shannon set forth his famous capacity theorem and in the process established the Holy Grail for communications engineers. Shannon's theorem proves the existence of a code that allows communications with arbitrarily small error rate over a band-limited channel in the presence of white Gaussian noise. Unfortunately, the Capacity Theorem offers no hint as to how such optimum codes can be found. In the ensuing years, progress in coding research has given us a host of coding techniques, including convolutional, block, and Reed-Solomon codes, and the mathematical theory with which to design codes with a desired level of performance. Such codes are regularly applied in everyday communications. In 1993, a new coding and decoding technique was described [ 11 that promised performance near the theoretical limit described by Shannon. Its authors dubbed the technique "Turbo Codes". The

. . . . . . . . . . . . . . . . . . .
Input Data Bits U Coder 1 output Parity Bits k=1.2,..m

vp

total times

!_._._ 4._.___ N-Point

.....................,

; lnterleaver j i Coderm j ; . . . . . . . . . . . . . . . . . . . . . : !. . . . . . . . . . . . . . . . . . . . . :

1.4 Systematic j __________ i


;

......................

i Desired

. . . . . . . . . . . . . . . . . . . . .

Figure 1: Turbo Encoder Block Diagram

0-7803-6395-7/00/$10.0002000 IEEE

8.B.2-1

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

output parity bits produced by each encoder are, in general, different. These parity bits may be or deleted to increase the code rate. The act of deleting parity bits is known as puncturing.

Iterative Decoding
The real "magic" in the Turbo Code concept comes in the iterative decoding process. Figure 2 illustrates this process for the case of two encoders, i.e., m=2 in Figure 1. The received signal y(O) corresponds to the data, and the received signals y") , and y"), correspond to the constituent parity bits. During any one source bit interval, three components (y'", y") , and y"), in this example) of the received signal enter the decoder on the left. The first operation is a maximum a posteriori (MAP) decoder matched to the first encoder. For Turbo Codes that use convolutional encoding, this MAP decoder is usually implemented by the BJCR algorithm described by Bahl [ 5 ] . A soft-decision Viterbi decoder may also be used, at some penalty in performance. Whatever implementation is chosen, the MAP decoding process uses the incoming data, the parity bits generated by the first decoder, and an estimate of the logarithm of the ratio of probability that each incoming bit is a "1" to the probability that the bit is a "0", (A,) The iteration is initialized by assuming ones and zeros are equally likely, and, therefore, A,=O.

For the second stage, the uncoded input data is reordered by means of the same interleaver used during the encoding process. The purpose for the use of systematic encoders, which provide an unmodified version of the input data, is now obvious. The reordered data, the second set of parity bits, and the updated estimate of the A, are now passed to the second MAP decoder, and the whole process is repeated. This decoding operation results in another update to the estimate of A,, and a second estimate of the received N-point data vector, i,. The process is repeated for each of the m constituent encoders represented in Figure 1. Once the process has completed the m decoding steps, we are ready for the next iteration. We first reorder the new A, by means of a deinterleaver so that it is in the original order and using it as the input to first MAP decoder. This provides new, and better, information about the distribution of "1"s and "0"s in the input data stream, y'". This iterative loop is continued multiple times. At the end of the total process, the final estimate, i,, is taken as the received data vector. As one might expect, the performance of the decoder increases with the number of iterations. To see this, we need only recognize that the performance at the end of the first half-iteration is the performance of the first constituent decoder by itself.

Initial

1 I Estimated 4.1 N-Point &=L4


MAP Decoder

Estimated A,

t-l Deinterleaver N-Point

Channel Capacity
The exciting feature of Turbo Coding is the ability of Turbo-Coded communications to approach the theoretical capacity of the physical channel. Shannon's Capacity Theorem states that there exists a channel capacity, C, and that errorfree communication is possible at any rate, R, such that ReC. Conversely, any attempt to communication at rates R > C will result in a probability of error of unity. A second important result, known as the Shannon-Hartley Theorem, gives a means of computing the channel capacity C for real-world channels corrupted by all-white Gaussian noise:

Y'u)(data$ Y"'(pan

b/

lnterleaver

MAP Decoder

discard

y") (parity)
N-Point lnterleaver

Figure 2: Block Diagram for Turbo Decoder with Two Constituent Encoders (m=2)
The output of the MAP Decoder is an updated estimate of A,, and an estimate of the received code. The estimate of the received code is discarded, and the second stage of the decoding process is initiated.

c = Blog,(l+-)N
3

8.B.2-2

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

where B is the channel bandwidth in Hertz, S is the signal power in Watts, and N is the noise power. In the limit as B + - ,

SB C=---log,e=1.44-,

S
No

where N o is the one-sided noise power spectral density.

Turbo Code Performance


One measure of the performance of a communications system is the characteristic curve relating the Bit Error Rate (BER) to the channel signal-to-noise power ratio (SNR). We can develop bounding curves shown in Figure 3 based on the average minimum distance between code words. The first curve is the upper bound (i.e. "worst possible performance") curve for a single convolutional encodeddecoder. The second is the asymptotic upper bound for the Turbo Code itself. A third curve may be represented by the ShannonHartley capacity. The result is a bounded area in SNR vs. BER space, as shown in Figure 3. For high and moderate values of BER, the Turbo Code performance will lie within the shaded triangular region. For low values of BER, the Turbo Code performance will lie below its asymptotic bound curve.
Shannon Capacity Limit
10 -O

design dictates a significantly lower BER, then conventional convolutional coding approaches may be more appropriate. For systems with end-to-end data verification, however, a channel BER in the range of lo-*to is generally acceptable.' In this case, Turbo codes may offer between 2 dB and 5 dB increase in performance over the forward error correction techniques currently used. The increased performance could be used to increase the information rate without increasing the bandwidth, or to maintain the same information rate in a more stressful interference environment, or to reduce radiated power requirements. All three types of performance gain would be welcome in any aeronautical application. The performance of a Turbo Code is driven by three closely linked factors: 1) the performance of underlying constituent encoder(s), 2) the number of iterations of the decoding algorithm, and 3) the depth and structure of the constituent interleaver(s). Choosing a good constituent encoder sets a lower threshold on the minimum distance and the asymptotic performance. Stronger encoders therefore create stronger Turbo Codes. The effects of the second and third factors are well documented, and can be summarized as follows: More iterations of the decoder algorithm give better performance; Deeper (longer) interleavers are better; These characteristics are illustrated qualitatively in Figure 4 and Figure 5, respectively.

,Bound

on Convolutional

e Performance Code Curve

I
-8

I B ound ' on Turbo Code Performance I

\ ,

Required Communications Performance


In recent years, the aeronautical communications community has been slowly moving toward standards based on the concept of Required Communications Performance (RCP). Under the RCP concept, civil aviation authorities will move toward certifying aircraft for operations based on the communications performance they can achieve
I Most aeronautical data links, including AMS(R)S, VDL Mode 2, VDL Mode 3, VDL Mode 4,and HFDL rely on a frame check sequence or cyclic redundancy check on each message to assure the integrity of the information passed to higher level entities.

Channel SNR (decibels)

Figure 3: Performance Bounds for Turbo Codes


The dashed line in Figure 3 qualitatively illustrates typical Turbo Code behavior. Turbo code performance curves are characterized by a region of rapidly improving performance with only small increase in SNR, followed by a flattening of the curve. This performance floor generally occurs in the vicinity of a BER of lo? If the system

8.B.2-3

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

through any means, rather than on the basis of carrying a specific suite of radios. Central to the entire concept is the specification of four critical parameters for each type of operational airspace. These parameters are Delay, Integrity, Availability and Continuity [6].

Delay and Integrity are both directly linked to the quality of the RF path between the user and satellite. In an effort to maximize the Integrity, that is, the probability that the user receives error-free data, most aeronautical data links implement some sort of Acknowledge-Repeat-Request (ARQ) protocol. The effect of this is to increase the Integrity at the expense of repeated transmissions of some or all of the information. Because no data are presented to the user until good data is available, \ these repeated transmissions increase the delay between when the data was sent from the source and when it was successfully delivered to the user. This integrity-based delay is, of course, random, based on the statistics of channel and the strength of the error detection implemented in the receiver. Other sources of random delay variations include 10-8 collisions caused by excess load on the communiI Channel SNR (decibels) cations channel, but this topic is beyond the scope of this paper. Figure 4: Qualitative Dependence of Turbo The effects of random integrity-based delays Code Performance on Decoder Iterations which require retransmission are amplified in the case of satellite communications by the path delay 10 -O associated with propagation distances that range N=1024 between hundreds and tens of thousands of nautical a , 10-2 miles. Application of coding techniques to reduce rcl the error rate by dramatically increasing the LI: 10-4 probability that the first transmission is error free, thereby eliminates costly (in the sense of additional w delay ) retransmission attempts. . = 10 -6 m Recognizing that Delay and Integrity go hand in hand, the remainder of this paper will examine the overall suitability of Turbo Codes for Channel SNR (decibels) aeronautical satellite communications, especially with a view toward high-integrity, safety oriented Figure 5: Qualitative Dependence of Turbo applications. For a standard of comparison, we will Code Performance on Interleaver Length consider communications via Inmarsat satellites in Generally speaking, Availability and near geostationary orbit, as specified in [7] and [SI. Continuity are related to the long-term and shortterm probabilities that a given a user can Transfer Delay Issues communicate over a given link. Although the error One of the key parameters defining the detection and correction techniques implemented on required performance for aeronautical commuthe link are certainly considerations in establishing nications is the transfer delay. [6] defines transfer a link's Availability and Continuity, other aspects of delay as "the period elapsed from the time at which the system design are significantly more important. the originating user initiates . .. [a communications] Therefore, the remainder of this paper will event until the time the transmitted information has concentrate on considerations Turbo Codes in term been received by the intended end user." Transfer of the RCP parameters Delay and Integrity. delay is one of the four defining characteristics for the Required Communications Performance (RCP)

e
L

8.B.2-4

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

that establishes what operations can be supported by a given communications link. Ninety-fifth percentile transfer delays may range from a few tenths of a second for digital voice applications to a few tens of seconds for data applications. Table 1 provides typical transfer delay requirements extracted from [7].

Broadband interference has a bandwidth greater than the matched filter processing bandwidth of the system under analysis. Therefore, some protection will be provided by the matched filter, which will reduce the total interference power by the ratio:

p = -B M F
W
where B,, is the matched filter processing bandwidth of the system under analysis and W is the bandwidth of the broad band interference. Any useful bandwidth definition can be used as long as the definition of BMFand Ware consistent. Potential sources of broad band interference to aeronautical satellite communications include broad band phase noise from other on-board transmitter or the broad band emissions from radars or communications devices. The intended emissions from the class of products collectively known as Ultra-Wide Band (UWB) emitters are of particular concern to the aeronautical communications, navigation and surveillance communities. Limits on these emissions, which can have bandwidths ranging up to several GHz, are currently under study by the FCC. Narrow band interference, also referred to as partial band interference, consists of interference that has significantly less bandwidth than the system under study. When this interference falls within the matched filter bandwidth, it has significant effects on the proper detection of the digital information. Narrow band interference also tends to have significantly higher levels than the received signal, potentially causing receiver desensitization effects. Narrow band interference that falls outside of the matched filter bandwidth is significantly attenuated. Potential sources of narrow band interference include harmonics of other communications systems operatin in lower spectrum bands. For example, the 13 harmonic of a VHF communication radio operating at 121.2 MHz falls within the GPS L1 frequency band at 1575.42 MHz.

Application
Digital Voice Safety-related digital data Flight regularity related data Non-safety data

95%-ile Delay
0.485 sec

10-20 sec

20-80 sec

110 sec

Multipath
For aeronautical applications, multipath is always present to some degree. The amplitude of multipath-induced fading is known to have a Rician distribution, to be sensitive to the characteristics of the Earths surface at the nominal specular reflection point, to be a direct function of the grazing angle with respect to the reflecting point, and to favor horizontal polarization. Hence, the carrier-to-multipath ratio experienced by an aeronautical terminal at any given time will be strongly influenced by its geographic location; its elevation angle to the satellite; the dynamic directivity and polarization characteristics of its antenna; the aircrafts maneuvers; and the current conditions of the Earths surface reflecting point. Typical fade margins are in the range 2 dB - 5 dB.

External Interference
With the increasing reliance on telecommunications technology for a wide variety of applications, interference among services is an increasing concern. This is especially true in the case of satellite-based services, where the received signal from the satellite can be on the order of -150 dBW. Due to the extreme difference in path loss, even well designed aeronautical communications, navigation, or surveillance equipment may pose interference threats. Interference sources can be generally grouped into three categories -- broad band, narrow band, and bursty.

Bursty interference may be either broad band or narrow band but is random in its occurrence with respect to the received signal. Bursty interference

8.B.2-5

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

tends to cause bursty errors, which must be either corrected or detected by the coding. Devices with relatively high peak power and low duty cycle, such as DME transponders and TCAS, present the potential for bursty interference.

Turbo Codes for Aero Applications


The final section of this paper looks at the high-level capabilities of Turbo Codes and assesses their applicability to the aeronautical SATCOM environment just described. This assessment is generally. qualitative in nature, as the quantitative details are subject of incomplete current research. We will also provide a high-level view of the applicability of Turbo Codes to a representative sample of potential aeronautical applications. For comparison purposes, we will select a baseline consisting of the signal format and protocols standardized by ICAO [8] for use with aeronautical safety communications via satellites. We will assume that communications are conducted on an Inmarsat 10,500 bps channel with encoding and as shown in Table 2.

standards already include interleaver depths that vary by factors of up to 150. Furthermore, the BER at the decoder output ranges from 0.001 to 0.00001, thus falling in the range where Turbo Codes offer the most advantage. Finally, the Rician multipath environment is relatively benign, offering us the possibility of obtaining performance bounds in a straightforward manner. To make the comparison as fair as possible, we would like to have a Turbo Code decoder that is equivalent in complexity to the K=7, R=1/2 convolutional encoder already implied by the standard. Let us assume that the standard decoder is a softdecision Viterbi decoder, and assume further that its complexity is proportional to the number of states, 2K-'. For our Turbo Code, we would like to use a MAP decoder based on the BCJR algorithm described in [5]. [9] provides an intuitive comparison between a soft decision Viterbi decoder and the equivalent BJCR, and concludes that, for the same constraint length, the BJCR decoder is no more than 4 times as complex as the Viterbi decoder. As described above, a half-rate Turbo Decoding algorithm requires two such decoders. To maintain a similar level of complexity, we need to choose our constituent code constraint, K , to satisfy:

System Parameter Channel Modulation Channel Modulation Rate Channel Coding Channel Interleaving

ICAO Standard Value


QPSK with RRC filtering 10,500 bps

2x2Krc-I
or, KTc = 4.

2K-' -

4 '
What this tells us is that by using two K=4, R=1/2 constituent encoders, and puncturing the results to obtain a R=1/2 Turbo Code, the complexity of a Turbo Decoder is upper bounded by the complexity of a soft decision Viterbi decoder for a straightforward K=7, R= 112 convolutional code. But this equivalent complexity has not yet achieved any performance enhancements, because we have not yet accounted for the fact that the Turbo Decoder must operate through multiple iterations to achieve its desired effect. To upper bound this effect, let us assume that the Viterbi implementation is such that it is strictly in step with incoming data. That is, let us assume that we consume 95.24 microseconds (1/10500 bps) for the soft Viterbi processing.

R=1/2, K=7 convolutional for data


R-channel: 64 x 5 block T-channel: 64 x 8 (min) 64 x 95 (max) P-channel: 64 x 78 block C-Channel: 5 x 8 block

Desired BER at decoder output CM, Required

R, S , T: c : 10.~ R, S , T: 44.3 dB-Hz

A comparison to this standard offers some insight into the tradeoffs necessary. From the discussions earlier in this paper, we know that Turbo Code performance increases with longer interleavers. Table 2 indicates that the current

8.B.2-6

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

Because we have designed the Turbo Decoder to be of equivalent complexity, we can assume that it also requires 95.24 ps for each iteration. A five iteration decoding process will then cost 476 ps, and a 20 iteration decoder will cost 1.9 ms. These times are small with respect to the length of the mean transfer delay given in Table 1; so, from that perspective, the additional processing required to achieve Turbo Code performance is quite acceptable. From an implementation standpoint, however, this may not be true. As we have currently structured our model, the Turbo Decoder will require between 5 and 20 bits of time to decode each incoming bit of information. The bursty nature of aeronautical data communications suggests that this excess processing time is not a "show stopper", provided that sufficient buffering is available to store incoming burst while the decoder plugs away. This factor is an artifact of our assumption of the underlying Viterbi decoding speed. Hardware decoding is, in general, far faster than we have assumed. Therefore, although our assumption, if truly representative of system performance, would impose limits on the average data transfer rate, it is useful as an upper bound on the time associated with Turbo Decoding.

Of the two categories, only ADS-C is applicable to SATCOM.~ A typical ADS-C message is on the order of 32 octets, or 256 bits. Thus we could use an interleaver of only 256 bits, which is short by Turbo Code standards. On the other hand, because ADSC information is generated only once every several seconds, iteration overhead is unlikely to be a problem. In the tradeoff between interleaver length and iterations, however, interleaver length is more desirable. Thus, a Turbo Code ADS-C application is unlikely to realize the full Eb/No performance advantages. With this cursory examination, we would conclude that ADS-C is not a likely candidate for Turbo Code implementation.

Distribution of Graphical Weather Data On the other hand, RTCA Special Committee 195 is currently completing work on a standard for transferring graphical weather data to the cockpit. This standard envisions application files of 100,000 bits or more. In this case, an interleaver length of 214 to 216 bits is certainly possible. This implies that significant performance advantages can be gained by a Turbo Code implementation.
But such gains do not come without cost. Returning to our worst case assumption of the Viterbi throughput, we see that Turbo Decoding of these long blocks could become the limiting factor in Weather Distribution. For example, consider an interleaver length of 216 bits with an R=1/2 Turbo Encoder on the Inmarsat channel defined in Table 2. Then we need to send 2" bits at 10,500 bits per second. This operation requires 12.5 seconds. If a five-iteration MAPS decoder is used, we could, under our worst case assumption, experience a decoding delay of nearly a minute. Fortunately, the SC 195 work anticipates that weather products, although large, will only be updated on intervals of five or ten minutes. Thus this worst case decoding delay is likely to be quite acceptable in terms of the weather distribution application.

Potential Operational Applications


Finally, we will take a brief look at potential Turbo encoded satellite communications from the viewpoint of particular aeronautical communications applications. We will consider four such applications -- automatic dependent surveillance, distribution of graphical weather information, controller-to-pilot-data link, and digital safety voice applications.

SATCOM ADS-C Automatic Dependent Surveillance (ADS) is the term applied to communication of position and flight path information from an aircraft to another entity or entities. There are basically two categories of ADS applications: ADS-B, which broadcasts the information to any user, including other aircraft, in the local airspace and ADS-C, which transmits the information to up to five specific entities such as Air Traffic controllers or airline dispatch offices.

The fact that we do not consider ADS-B in this paper does not mean that Turbo Code techniques might not be applicable to ADS-B, but only that ADS-B is not a likely candidate for. SATCOM applications and is therefore outside the scope of this paper.

8.B.2-7

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

With this cursory examination, we would conclude that distribution of graphical weather is a likely candidate for Turbo Code implementation

SATCOM Voice
Finally, we will consider use of Turbo Coding techniques to improve the ICAO standard voice channeL3 Turbo Coding for voice applications is already widespread in the digital cellular telephony applications, and has been included in the next generation digital cellular standard. The downfall for Turbo Code operation in a channel operated in accordance with the current ICAO SATCOM voice standard is the very limited interleaver depth of 40 bits. With an interleaver this small, the ability of the Turbo Encoder to randomize the inputs to the constituent encoders is severely limited. This forces the Turbo Code performance bound higher and effectively wipes out any performance advantage. Furthermore, under our worst-case assumptions concerning MAP processing, the decoding process might induce an additional 1015% delay in the voice channel. Such a worst case performance is likely to be unacceptable. Therefore, we conclude that under the constraints imposed by the current ICAO standards, Turbo Coded SATCOM voice is unlikely to offer any significant advantage. Two final caveats are in order as a postscript to this entire section. First, for the purposes of illustrating the high level tradeoffs to be considered, we have focused on a well documented and standardized communications link defined in [8], and have imposed limitations on our analysis to maintain some level of comparability with current implementations. These limitations may not be imposed on real Turbo Coded applications. For example, ICAO has recently approved a new chapter establishing generic standards for "Next Generation Satellite Systems". Systems complying with this chapter need not comply with Chapter 4 [8]. Thus, the choice of data rate, modulation, interleaver, etc., will again be open for system performance tradeoff.

Second, we have made an extreme assumption regarding MAP processing time. Hardware Viterbi decoders are widely available and run at VLSI speeds that far exceed the channel bit rate considered in this example. Viterbi [9] concluded that there are no significant obstacles to hardwarebased MAP decoding at many megabits per second. Recent Turbo Code research, to be published in September 2000, reports on several hardware-based Turbo Code implementations. This paper has provided a very brief and very high level overview of Turbo Codes and considered some examples of how they might be applied to aeronautical SATCOM. Turbo Codes are real: they are being applied in commercial telecommunications around the world. The advent of fast, hardware-based decoders is likely to extend the range of Turbo Code applications into higher bandwidth data communications. The information provided in this paper was developed during course work and research activity at University of Maryland, Baltimore County. Participation in these activities was partially supported by the Honeywell continuing education program.

References:
[ 11 Berrou, C., A. Glavieux, and Thitimajshima, "Near Shannon Limit
Int. Con& on Comm., Geneva, Switzerland, pp

Error-correctingCoding and Decoding: Turbo Codes, Proc. 1999 IEEE 1064-1070, 1993.
and

[2] Vucetic, B., J. Yuan, Turbo Codes Principles Applications,Boston: Kluwer Academic Publishers, 2000

131 Hagenaur, J. "Iterative Decoding of Binary Block and Convolutional Codes", Transactions on Information Theory, Vol. 42, No. 2. IEEE, March 1996 [4] Sklar, B., "A Primer on Turbo Code Concepts", Communications Magazine, . IEEE, January 1997

IS] Bahl, L. R., J. Cocke, F. Jelinek, J. Raviv. "Optimal Decoding of Linear Codes for Minimizing Symbol Error Rate". IEEE Trans. Inform. Theory, Vol. IT-20, March 1979
[6] SC-189NG-53, Guidelines for Approval of the Provision and Use of Air Traflc Services Supported by Data Communications, RTCA Paper 199-00/SC189-038,July 2000. 171 SC-165, DO-210D Minimum Operational Performance Standards for Aeronautical Mobile Satellite Services (AMSS), Washington: RTCA, Inc. ,2000 [8] ICAO, "Aeronautical Mobile Satellite Service", Annex 10 to the Convention on International Civil Aviation, Volume 111. Part I , Chapter 4. Montreal: Intemational Civil Aviation Organization

In March 2000, ICAO approved a second voice standard using different parameters and operating at a lower vocoder rate. This example is based solely on the older 9600 bps Inmarsat C-Channel information contained in the ICAO standard prior to the recent changes.

[9] Viterbi, A. J., "An intuitive justification and a simplified implementation of the MAP decoder for convolutional codes", IEEE Journal on Selected Areas in Communications, vol. 16, no.2, p. 260-4 Feb 1998.

8.B.2-8

Authorized licensed use limited to: Florida Institute of Technology. Downloaded on April 4, 2009 at 06:06 from IEEE Xplore. Restrictions apply.

Vous aimerez peut-être aussi