A Feedback Adaptive Variable Rate Meteor Burst Commuekatlon System.

Sheldon S.L. Chang (Fellow, IEEE)

Department of Electrical Engineering State University of New York at Stony Brook

Stony Brook, N.Y. 11794

Abstract

Meteor burst communication (MBC) channels are known for their intermittant nature and large variations of received signal power. For optimum utilization of such channels, a feedback adaptive variable rate (FAVR) system is proposed in which the transmitted bit rate is proportional to channel transmittance (or received signal power). Based on well accepted channel models for low density meteor bursts, the expected improvement in throughput is significant: approximately five times that of constant bit rate systems, and ten times that of constant bit rate systems with ARQ. The system adapts automatically to give optimum throughput for both high and low density bursts. Considering the occasional presence of high density bursts, the improvement of throughput over a con- stant rate system is even higher.

Two other advantages are: (i) I t requires minimum waiting time among MBC systems, (ii) the transmitted bits can be readily formulated into packets with only minimal loss of the throughput rate. Thus the FAVR-MBC systems can be readily incorporated into any existing communication net- work.

The FAVR system operates on the physical principle that a MBC channel is reciprocal. The channel transmittance in one direction is an approximate but close measure of its transmittance in the opposite direction. Required terminal equip- ments for realizing the FAVR system are described.

Introduction

The meteor burst communication channel offers an alternative means of communication which has scarcely been utilized. Yet it has some interesting advantages over other beyond-line-of- sight systems. These include Iower susceptibility to

atmospheric perturbations and interference, low cost, and relative safety from destruction, [I, 21. However, IOW throughput rate and its intermittent nature are among the factors which retarded its development. The present paper discusses a method which alleviates these two factors substan- tially, and thereby makes the meteor burst channel more feasible as a practical means of communica- tion.

Presently known meteor burst communication systems includes the JANET system 131, the COMET system [4], and a modified COMET sys- tem proposed by Booz-Allen [SI. In the JANET system, which uses Amplitude Modulated (AM) sig- nals, the gating is performed by monitoring the Signal-to-Noise Ratio (SNR) at the input to the system and the information rate is constant. The COMET system, which uses a Frequency Shift Keyed (FSK) modulation, performs the gating by an Automatic Request (ARQ) when an error is detected. A similar technique is used with the Booz-Allen scheme using ARQ and Hybrid ARQ/FEC (Forward Error Correction). Again, in both these systems a constant rate of transmission is used and presumed to be optimum.

The Feedback Adaptive Variable Rate (FAVR) system denotes a specific method of realizing a variable bit rate system in which the energy per bit, or bit error probability is kept constant. While the idea of a variable bit rate system is not new in the literature 16, 71, it has been regarded as difficult to implement. The present paper gives an analysis of the advantages to be gained with a variable bit rate system. Then it gives a description of the FAVR system.

Analytical Comparison of Three Types of Systems

In the following, the throughputs of three

* This work is sponsored by the National Science Foundation under award number ISI-8660079 to SCS Telecom, Inc.. Dr. Chang is a consultant to SCS Telecom, Inc..

14.2.1. CH2538-718810000-0423 $1 .OO 0 1988 IEEE 0423

types of MBC systems are compared. To make the comparison meaningful, the same channel model is used for all three types of systems. The channel model is described as follows [8,9]:

1. The received signal power during one meteor burst can be represented as

p = p , e - ' l B (1) where P,,, is the peak power, and B is a time con- stant.

P , and B are random variables for the ensemble of meteor bursts. For each indivi- dual burst, P, and B are constants. The dis- tributions of P , and B are independent:

f(a) = average number of meteor bursts per second with P , 2 a . PB ( 6 ) = probability of B 2 b.

The occurrences of meteor bursts are Poisson.

The frequency function f( a) can be expressed as:

f ( a ) = Kl(a -" - P i " ) , 0 < U 2 P i (2)

where P , is the maximum value of the peak power, and f(a) = 0 for a > P 1 . Equation (2) gives a probability density function

p ( a ) = C U - ( ~ + ' ) , f o r a 5 P , = O f o r a > P ,

which has been well documented in the literature for low density bursts [8, 9). The frequency func- tion is directly measurable: f(a) is the average number of meteor bursts per second with peak power P, 2 a .

The mathematical model presented above is for the purpose of comparing the three types of sys- tems. It is not essential to the FAVR system's operation.

Let E denote the received signal energy per bit for a prescribed signal to noise ratio (e.g. 10dB). Let p denote the bit rate, and Po denote a lowest designated pulse power. Then

Po = p E (3)

for a constant bit rate system. For a variable bit rate systeri, (3) is replaced by

Po = P o E (3A)

where po is the lowest bit rate of the system. For both types of systems transmission ceases when P < P o .

I. Constant bit rate svstem

For a constant bit rate system p is a constant. The usable duration t , is determined from (1):

a t , = b In - PO

The number of bits in a pulse is

(4) a Nc ( a , b ) = p t , = p b In - PO

The throughput rate (average bits per second) is 00 Pl

0 P,

Integrating with respect to a gives p1 p1

P O

Nc = s f N c ( a , b ) df (4 d P B ( b ) (5)

-1 N , ( a , b ) df ( a ) = f f (4 d W a , b )

Substituting the above into (5) gives

where 2 = P o / P l . The specific throuEhDut f , ( n , z ) i s

A plot of f (0.6, x) versus x is shown in Fig. 1A.

II. Jdeal feedback adaDtive svstem:

The bit rate p varies inversely with P, and the pulse energy E is constant. The number of bits in a meteor burst is

t o

(8) 1 N j ( a , b ) = - S P dt E o

where to is given by

14.2.2. 0424

- t o l b Po = a e f

N , ( a , b ) = J dt E O

b = -(a E - P o )

The throughput rate is 00 P1

0 Po N/ = j j Nf ( a , b ) df ( a ) d P B ( b )

Integrating the above expression gives:

f & ( n , r , z) = -[x-" r x e-' -11 (18) (9) n

To maximize transmission, the optimum values of r and x are determined as follows:

(i)

-- -0 af Q a2

(11) (ii)

-- -0 af 9 dr

where f b (n, x) is the specific throughput function: c

A plot of f (n, x) versus x is shown in Fig. 1B.

III. ARQ svsteq

In an ARQ system, a packet of N bits is transmitted at the detection of each meteor burst. For any a and b, let P 2 be defined by

(14) p2 PO

p b In- = N

(20)

ro = n (21)

( l - n ) n (22)

2 - n e-' -1-r 2 - n e-' = 0

Solving from (19) and (20) gives

1 - 1 2, = -

e

The maximum value of f is 1-n -

(23) n e

Substituting (21) into (16) gives the optimum packet length for an ARQ system:

f &mu = - (1-n)

No =-= ' O b p b E (24)

For n = 0.6 we obtain from (22) and (23):

X, = 0.0799 (25) f QmW = 0.120

decay time b:

N g ( b ) = N f

= K I N

nN Let r = - P b

Eq. (15) can be

instead of b in (17) and (24). Consequently

( P z ) = N K 1 (P;" - P i " ) f Qm(u < 0.120 (26)

N = p B (27)

-nN with

. Then

Po br nE

N = -

written as

A plot of f Q ( n , r , x) with r = n is shown as Fig. 1C.

The following conclusions can be drawn from (16)

the above analysis:

K l b NQ ( b ) = E P;-" f (n , r , 3 ) (17) 1. The throughput of a feedback adaptive system

is significantly higher than that of a constant bit rate system, including the ARQ system. where the specific throughput is

14.2.3. 0425

~

2.

3.

The optimum value of Po depends on the sys- tem. A feedback adaptive system can utilize low density meteor bursts to better advantage.

The FAVR system requires minimum waiting time. We note that while ARQ systems depend on a few large meteor bursts, which are few and far in between, feedback adaptive systems utilizes every meteor burst that can be utilized. Its waiting time is the least possible with the meteor burst communication channel.

Realization of the FAVR System The FAVR system is realized by computer

controlled transmission and reception with FSK or BPSK modulation. We shall describe the FAVR system in the following sequence:

1. Signal composition 2. Transmitter 3. Receiver

Siena1 comDositioq

Each bit of signal is made up from an integer number f l b of signal chips. The duration of a chip is a constant of the system. It is denoted as T, . The bit transmission time is changed by varying f l b . In the proposed system, f l b is increased by multi- plying with factors 312, and 4/3 alternatively: 2, 3, 4, 6, 8, 12 etc.

There are n, bits in a packet segment (PS), and an integer number of packet segments in a packet. The number f l b remains constant in a packet segment. The number nr is a constant of the system. During a meteor burst an integer number of PS are transmitted. The number varies with the strength of the meteor burst. Thus a mes- sage packet can be transmitted in a number of meteor bursts, and a meteor burst can transmit a number of message packets depending on its strength.

Transmitter

Figure 2 is a block diagram of the FAVR transmitter. During the standby period, the transmitter a t each end sends out square wave sig- nals of half period equal to T l .

At the arrival of a meteor burst, the

handshake processor at each receiver picks up a strong square wave signal. The FAVR system operates on the basis of reciprocity of the meteor burst channel: that assuming identical terminal equipments a t the two terminals A and B, the received signal strength from A to B is a close approximation of the received signal strength from B to A. From the received square wave, the handshake processor determines the required bit duration for a specified signal to noise ratio. Usu- ally the required bit duration falls between two allowed values of f l b T,. The larger f l b is then selected, and transmission begins.

The channel condition is monitored through the receiver signal to noise ratio. When the latter falls below a critical level, the Bit Timing Control signals for the next lower rate of transmission or next larger f lb value. The transmitter then transmits a t the new rate starting with the next packet segment without signaling the receiver, which would then follow the change of rate automatically, as described in the receiver section.

Receiver

A block diagram of a FSK receiver is illus- trated in Fig. 3. It has a simple front end analog circuit which yields four numbers for each signal element: z l i , y l i , z2i , and y2i . These are the cosine and sine Fourier components in the received signal at the two alternative frequencies q a n d w, respectively.

Five microprocessors operate in multi- processing to yield the detected signal and an instantaneous best estimate of the signal to noise ratio.

Large values of any of the four numbers alert the Handshake Processor which computes the sig- nal to noise ratio and determines the initial f l b to be used. This information is transmitted to the transmitter and also to the Central Track Proces- sor. The two other track processors are then set at next higher and lower values of nb respectively.

The average signal to noise ratio over a packet segment is computed within each track. There is a track buffer within each track which stores 2 n, bits of information, and the demodulated bits are temporarily stored in the track buffer. The Control Processor compares the average signal to noise

14.2.4. 0426

ratios of the three tracks and forward only the stored signal with highest SNR to the Main Buffer.

Simal Processine: for Variable Bit Rate FSK

Variable bit rate FSK is realized by using digi- tal processing of the detected signal as illustrated in Fig. 4. We assume that within the short transmission time of 1 bit the received signal phase is essentially constant.

As discussed previously, the time tb for transmitting 1 bit can be expressed as an integer multiple of T :

tb = nb TI (31)

Let r(t) denote the received signal, and w l , w2 denote the alternative FSK frequencies.

i TI

21i = s r ( t

Y1i = S r ( t

(i-l)T, i T I

(i-1) T

The variables x2i and y2;

sinwlt dt (33)

are similarly defined with w2 replacing w1 . For each bit of transmitted signal, i runs from 1 to nb :

t -n(

i = l i a n ,

i-1

X I = 21i . (34)

Y I = C Y1i (35)

The variables z 2 , and y 2 are similarly defined. Let

z 1 = z 1 + Y 1

2 2 = 2 2 + Y2

If z l > z 2 , then the transmitted signal is at w1 . The above system is readily shown to be unsyn- chronized FSK. However its bit rate can be changed by changing nb in (34) and (35).

(36)

(37)

2 2

2 2

Conclusion

The Feedback Adaptive Variable-bit-Rate Sys- tem is a means for maximizing the throughput, and minimizing the waiting time of a meteor burst com- munication channel. Its packet segmentation struc- ture makes it usable as either a link in a meteor burst communication network or an additional link in a conventional communication network.

There is a some discussion in the MBC litera- ture on whether to use high bit rate to utilize the infrequent high density bursts, or to use low bit rate to utilize all meteor bursts. The FAVR system settles this issue by automatically adapting the bit rate to the transmittance of the channel.

Three important technical problems must be solved as prerequisite conditions for realizing thr variable bit rate concept: (i) a variable bandwidtl filter or the equivalence of it, (ii) accurate and timely determination of the signal to noise ratio, and (iii) communication of the bit rate information. The three problems are solved in the FAVR system by using a receiver with a simple analog circuit fol- lowed by digital processing of the analog signals. The variable bandwidth filter problem is solved by summing a variable number of terms of the analog signals. Signal power and noise power are deter- mined separately by using Kalman’s optimum filtering and predication algorithm. Changes in bit rate are not explicitly communicated, but deter- mined by the receiver digital circuit.

The FAVR system is an example of using digi- tal processing extensively in optimizing the perfor- mance of a difficult communication system. Digital processing, once developed, is reliable, inexpensive, and flexible. With the current trend of better and cheaper chips every year, the role of digital process- ing in communication is expected to expand and to grow.

References

[l] G.R. Sugar, “Radio Propagation by Reflection from Meteor Trails”, Proc. IEEE, Vol. 52, pp. 116-136, Feb. 1964.

121 J.D. Oetting, “An Analysis of Meteor Burst Communications for Military Applications”, IEEE Trans. Commun., Vol. COM-28, pp. 1591-1601, Sept. 1980.

[3] P.A. Forsyth, E.L. Vogan, D.R. Hansen, and C.O. Hines, “The Principles of JANET - A Meteor Burst Communications Systems”, Proc. IRE, Dec. 1957.

[4] P.J. Bartholome and I.M. Vogt, “COMET - A Meteor Burst System Incorporating ARQ and Diversity Reception”, IEEE Trans. Commun. Technol., Vol. COM-16, April 1968.

14.2.5. 0427

J. Hampton, “A Meteor Burst Model with Time Varying Bit Error Rate”, MILCOM 1985, 32.2.1.

L.L. Campbell and C.O. Hines, “Bandwidth Considerations in a JANET System”, Proc. of IRE, Vol. 45, pp. 1658-1661, Dec. 1957.

E. Hibshoosh, D.L. Schilling, “Time Varying Bit Error Rate for Meteor Burst Channel”, MILCOM ’86, Monterey, CA., Oct. 1986.

V.R. Eshleman and L.A. Manning, “Radio Communications by Scattering from Meteoric Ionization”, Proc. IRE, Vol. 42, pp. 53G536, March 1954.

C.O. Hines and P.A. Forsyth, “The Forward Scattering of Radio Waves from Overdense Meteor Trails”, Canad. J. Physics, Vol. 35, Feb. 1957.

STANDBY 8 HANDSHAKE SIGNAL GENERATOR

MESSAGE TRANSMITTER

BIT RATE CONTROL

a

m

FROM RECEIVER

Fig. 2 FAVR Transmitter control signal

t- 3 a I (3 3

I I- W > l-

J w K

z - a

’ 0.01 0.02 0.05 0.1 0.2 0.5

x = Po / P, Fig. 1 Throughputs of three types of communication systems

14.2.6. 0428

r (t) ANALOG

A TO D CIRCUITS

. AND -.c

UPPER TRACK

CENTRAL

PROCESSOR

L r CONTROL MAIN OUTPUT - TRACK -

PROCESSOR BUFFER P

__c

LOWER TRACK

PROCESSOR

-

- TO CENTRAL TRACK PROCESSOR (FOR INITIAL nb) - GATING SIGNAL TO TRANSMITTER - lPRoCESSoR -TO TRANSMITTER INITIAL B I T RATE

"b FROM CONTROL PROCESSOR

cos w, t T I I I

I I I 1 8 I

S W I T C H X,' SQ 1

I n

TO CONTROL

PROCESSOR

MESSAGE OUTPUT -c

I I I

I AND I TYPICAL TRACK PROCESSOR

ANALOG

A TO D CIRCUITS

Fig. 4 Signal formation in Track Processor

14.2.7. 0429