CA1157121A - Effective error control scheme for satellite communications - Google Patents

Abstract

AN EFFECTIVE ERROR CONTROL SCHEME
FOR SATELLITE COMMUNICATIONS

ABSTRACT

A hybrid scheme for controlling transmission errors in digital data communication systems. Normally, data blocks with a small number of parity digits for error detection are transmitted. When the presence of errors is detected, the retransmissions are not the original data blocks but some properly selected blocks for correcting errors in those erroneously received data blocks which are stored in a buffer at the re-ceiver. The retransmitted blocks are formed based on the original data blocks and error-correcting codes with an invertible property. When such blocks are received, they are used to recover the original data blocks either by an inversion operation or by a de-coding process.

Description

~ YO979-06~

AN EFFECTIVE ERROR CONTROL SCHEME
FOR SATELLITE COl`~UNICATIONS

DESCRIPTION
-Background of the Invention The present invention relates to digital data communi-cation systems and more particularly to a technique for controlling transmission errors in such systems.

There are basically two known categories of schemes for controlling transmission errors in data communica-tion systems: the automatic-repeat-request ~ARQ) scheme and the forward-error-correction (FEC) scheme.
The ARQ scheme is widely used because it is simple and it provides high system reliability. ~owever, systems using ARQ scheme have a severe drawback, their throughput falls rapidly with increasing channel error rate. For channels such as satellite channels with large round trip delay, this problem becomes even worse. Systems using the FEC scheme maintain a con-stant throughput regardless of the channel error rate.
However, the FEC scheme is less reliable than the ARQ
scheme. Moreover, it is very difficult to implement when a large collection of error patterns is to be corrected.

There are basically three types of ARQ schemes, these ~eing the Stop-and-Wait, Go-Back-N, and Selective Repeat. In any of these three techr.iques, the re-ceiver sends the sender;the results of the error detection in Dj,C(Di). If no error is detected, a positive acknowledgement (ACK) is sent. Otherwise, a negative acknowledgement (NAK) is sent. The three techniques differ in what the receiver and sçnder do between the instan~ the sender has completed trans-.~.~, ~V979-06G

mitting Dj and the instant the sender receives the acknowledgement for Dj. In Stop-and-Wait, nothing is don~ in that period. In selective reject, with a limited number of outstanding packets allowed, trans-mission of new packets and/or repeat of the packetsfor which a NAK has been received may occur in that period. In Go-Back-N, the N-l packets which follow Dj are transmitted in that period. Besides the differ-ence described above, these techniques may differ in the length of the interval between the instant a NAK
is received (if an error is detected in Dj,C(Dj) for Dj and the instant Dj is repeated.

Although the Stop-and-Wait ARQ is simple, it is in-herently inefficient due to the idle time spent wait-ing for an acknowledgement for each transmitted datablock. This inefficiency becomes unacceptable for systems where transmission rate is high and xound trip propagation delay is large, such as in satellite channels. Selective Repeat ARQ is the most efficient scheme among the three types, but has implementation problems such as infinite buffer requirement or buffer overflow, if a buffer of finite size is used. The ~o-Back-N ARQ, while being less efficient than the Selective Repeat ARQ, is more efficient than the Stop-and-Wait ARQ and its implementation is simpler than that of the Selective Repeat ~RQ.

In a Go-Back-N ARQ system, the transmitter sends data blocks continuously to the receiver and the receiver sends acknowledgements to the transmitter continuously.
When a data block is received successfully, a positive acknowledgement (ACK) is sent to the transmitter.
However, when a data block is detected in error, a negative ac~nowledgement ~NAK) is sent to the trans-mitter. On receiving a NAK, the transmitter backs up to the data block that was detected in error at the ~Si ~,1 receiver and retransmits that data block and subsequent data blocks transmitted in the interval between the original transmission and the receipt of the NAK. At the receiver, the data blocks following the erroneously received data blocks are discarded no matter whether they are received correctly or not. This scheme is simple. However, since it usually involved retrans-mitting good data blocks following an erroneous data block, its efficiency, the maximum achievable through-put, becomes unacceptably low when channel error rateis high and round trip propagation is large.

In a Go-Back-N system, after the last data block has been transmitted, the transmitter stays idle until arrival of the next data block or a NAK of a previously transmitted data block. In a recently proposed Go-Back-N scheme, during the period of time when the transmitter would be normally idle under the Go-Back-N scheme, it repeatedly transmits the last data block.
Although such Go-Back-N scheme improves the response time from the data block entering the transmitter until it being output by the receiver, its efficiency is still the same as the basic Go-Back-N ARQ.

It is an object of the present invention to control transmission errors in data communication systems in a simple and efficient manner.
.
It is another object to control transmission errors in a mo~e efficient and reliable manner than otherwise achieved with the known automatic repeat request (ARQ) and forward error correction (FEC) schemes.
. ~
It is a further object to improve transmission effi-ciency when a data communication channel is very noisy and the round trip propagation delay is large.

Summary of the Invention These and other objects a_e achieved by the present invention which provides a method for controlling transmission errors in digital data communication 5 systems including transmitting data blocks with a small number of parity bits for error detection and, when the presence of errors is detected, retransmitting properly selected parity blocks for correcting errors in the erroneously received data blocks which are stored in a buffer at the receiver. The parity blocks are formed based on a half-rate invertible code. A
code is said to be invertible if, knowing only the parity digits of a code word, the corresponding in-formation digits can be uniquely determined by an inversion process. Hence, if the parity blocks are received correctly during the retransmission cycle, the original data blocks can be recovered by a simple inversion process without going through a more compli-cated decoding process. However, if the parity blocks are not received successfully, they will be used to correct errors in the erroneous data blocks previously received. Using this method, high system efficiency and reliability can be achieved even when the channel error rate is high and the round trip propagation delay is large.

Brief Description of the Drawings FIG. 1 shows the data blocks at a transmitter and a receiver for a conventional automatic-repeat-request ~ARQ) scheme.

FIG. 2 shows the data block arran~ement at a trans-mitter and a receiver for the selective parity re-transmission (SPREC) scheme in accordance with the present invention.

Y097~-066 FIGS. 3 and 4 show a data and parity block arrangement at a transmitter and a receiver where two retrans-mission cycles are involved in the scheme of the present invention.

FIG. 5 is a system block diagram of the transmitter of the present invention.

FIG. 6 is a system flow diagram for the transmitter shown in FIG. 5.

FIG. 7 is a system block diagram of the receiver of the present invention.

FIG. 8 is a flow diagram corresponding to the opera-tion of the receiver shown in FIG. 7.
.

FIG. 9 is a system flow diagram corresponding to the control program for the transmitter.

FIGS. lOA and 10B constitute a system flow diagram corresponding to the control program for the receiver.

FIGS. llA and llB are the data and parity block - arrangement at a transmitter using the selective parity retransmission (SPREC) scheme, illustrating the recovery period for an erroneously received data block. -FIGS. 12 and 13 are graphs showing the plots of th~
throughput of the SPREC system, and the throughputs of the conventional automatic-repeat-request (ARQ) schem~s for comparison purposes.

~'0979-066 Description of the Preferxed Embodiments In order to simplify the understanding of the subject invention, there will first be described the conven-tional Go-Back-N ARQ illustrated in FIG. 1. Using the conventional Go-Back-N ARQ, the transmitter sends data blocks 1 continuously to the receiver and the receiver sends acknowledgements to the transmitter continuously. When a data block 2 is received successfully, a positive acknowledgement (ACK) is sent to the transmitter. However, when a data block is detected in error, a negative acknowledgement (NA~) is sent to the transmitter. On receiving a NAK, the transmitter backs up to the data block that was de-tected in error at the receiver and retransmits that data block and the subsequent N-l data blocks trans-mitted in the interval between the original trans-mission and the receipt of the NAK as in FIG. 1. In the FIG. 1 example, N=7. At the receiver, the N-l data blocks following the erroneously received data block are discarded no matter whether they are re-ceived correctly or not. This scheme usually in-volves retransmitting good data blocks following an erroneous data block, and becomes very inefficient when channel error rate is high and the round trip delay is large.

The performance of an error control scheme is measured by its throughput and reliability ~or error probability).
The throughput is defined as the ratio of the average number of data blocks accepted by the receiver per unit of time to the number of data blocks that could be transmitted per unit`of time if no transmission errors occur. Let p be the probability that a data block is successfully received. The probability p consists of two parts, the probability Pc that a data block is received without errors and the probability Yos7s-066 Pe that a data block is received with undetectable errors i.e., p = Pe + Pc The throughput of the con-ventional Go-Back-N ARQ is given as the following:

~GBN P+(l-p)N (1) It is noted that the feedback channel is assumed to be noiseless and the overhead due to the parity bits of each block is ignored. The reliability of an ARQ
scheme is measured by the probability that the re-ceiver accepts a data block with undetectable errors.
Let PARQ(~) denote this error probability. Then, we have PARQ(~) p +p . (2) The probability Pe depends on the error-detecting code being used in the ARQ system. If an (n,k) linear code is used for a binary symmetric channel (BSC) with transition probability ~, Pe is upper bounded by P~ < 2 (n-k)~ n]

Generai1y, 2 (n k) is used as an upper bound on Pe for both channels with random or burst errors. If the error detecting code is chosen properly, ~e can be made very small. Consequently, the error prob-ability PARQ(~) becomes very small.

The su~3ect inventio~ pro~ides using a hybrid ARQ
technique using selective parity retransmissions hereinafter referred to as SPREC. The SPREC technique employs two codes, C0 and C1. The code C0 is a high rate (n,k) cyclic (or shortened cyclic) code which is used for error detection only. Such a code is used in every type of ARQ for checking the validity of ` yo979-066 data blocks and can be implemented easily with a feedback shift register. Both cyclic and polynomial codes are described in U.S. Patent 3,872,430 issued on March 18, 1975 to P. E. Boudreau et al. The second code Cl is an invertible half rate (2k,k) code which is capable of correcting t or fewer transmission errors and simultaneously detecting d(d>t) or fewer errors. For bursty channel, Cl would be a burst-error correcting and detecting eode.

A data block, denoted by (D,Q) consists of two parts, D and Q. The part D consists of k information digits and the part Q consists of n-k parity digits which are formed based on the information block D and the (n,k) code C0. Thus, (D,Q~ is a code word in C0. When a data block is received, the parity digit3 are used to eheck the validity of the block (i.e., error deteetion).

When a received data block, denoted by ~D,Q), is detected in error, the receiver requests for a re-transmission. The erroneous information block D is saved in a b~ffer. The retransmission is not the original data bloek (D,Q) but a parity block of n digits, denoted by (P(D),Q(l)), which consists of two - parts, P(D) and Q(l). The first part P(D) consists of k parity digits whieh are formed based on the original k information digits D and the half rate (2k,k) inverti~le eode Cl. Hence, (D,P(D)) is a code word in Cl. The second part Q(l) eonsists of n-k parity digits ehecking on P(D). These n-k parity digits are formed based on P(D), regarded as k infor-mation digits, and the (n,k1 code C0. Clearly(P(D),Q(1)) is code word ln C0.

When a parity block is reeeived, denoted ~y (P(D),Q(l)), the part Q(l) is used to eheck the validity of P(D).
If no errors being detected, P~D) is used to recover Yo~7s-0~6 the original data block D by an inversion process.
If the presence of errors is detected, then P(D) and the previously received data block D together are used for error correction based on the decoding rule of code Cl. If the total number of errors D and P(D) does not exceed t, the errors in D will be corrected.
However, if the number of errors in D and P(D) exceeds t but does not exceed d, errors will be detected. In this case, another retransmission is requested. The retransmission is again a parity block (P(D),Q(l)).
Retransmissions continue until the original data D is recovered either by inversion or by a decoding opera-tion.

In the description of the selective parity retrans-mission (SPREC) scheme of the present invention, for the convenience of presentation, the n-k error detect-ing parity digits is dropped from each data block and from each parity block. That is, (D,Q) and (P(D),Q(l)) are represented by D and P(D) respectively. The SPREC
scheme is illustrated in FIG. 2.

In a normal operation, the transmitter continuously sends data blocks to the receiver. The receiver checks the validity of each incoming data block based on error detecting code C0, and sends an acknowledge-~5 ment to the transmitter for each received block. Whenthe channel is ~uiet, the transmission proceeds smoothly and error-free data blocks are delivered to the user continuously.

When the first received data block Dj is detected in error r a NAK(i) is sent to the transmitter and the erroneous data block Dj is saved in a buffer. The receiver proceeds to check the validity of the sub-sequent N-l received data blocks and stores them in the buffer. Upon receiving the NAK(j), the trans-~Q979-0~6 mitter stops transmitting new data blocks, it baeks up N blocks to ~j (i.e., Go-Back-N) and starts retrans-missions. In the FIG. ~ embodiment, N=7. The re-transmissions are not the N original data blocks Dj,Dj+l,...,Dj+N_l, but the parity block P(Dj) and N-l properly selected parity blocks. The selection for retransmission is carried out in the following manner. For j<Q<j+N, if the data block DQ is not successfully received by the reeeiver, the transmitter sends the parity block P(DQl upon receiving the NAK(~).
If DQ is successfully reeeived by the reeeiver, the transmitter, upon reeeiving ACK(Q), does not send P(DQ) but uses the time slot to repeat the parity block P(Dj). The main idea is to use the time slots eorresponding to the successfully received data blocks following the first erroneously received data block Dj to send the parity block P(Dj). This transmission of P(Dj) and N-l properly selected parity blocks is - referred to as the first retransmission cyele. When these parity blocks are received, they are used to reeover Dj and other data bloeks erroneously recei~ed.
In FIG. 2, data bloeks 3, 7, and 9 are recovered as shown.

If the reeeiver fails to reeover Dj from the first parity bloek P(Dj) transmitted in the first retrans-mission cycle, it notifies the transmitter to start the second retransmission eyele. The first bloek in the second retransmission eyele is again the parity block P(Dj). The subsequent retransmitted blocks are parity blocks selected in a proper manner. The second retransmission cyele will be terminated as soon as the transmitter is notified that Dj has been re-co~ered by a parity bloek P(Dj) transmitted in the first retransmission cycle. ~he parity bloeks in the second retransmission eyele are seleeted in the follow-ing manner. If a parity P(Dj) transmitted in the ~'O9/9-066 7~ '1 first retransmission cycle fails to recover Dj, the transmitter, up on receiving a notification (NAK(j)), repeats P(Dj) in the second retransmission cycle, as illustrated in FIG. 3. For j<Q<j+N, if an erroneously received data block DQ is not successfully recovered by its parity block P(DQ) transmitted in the first retransmission cycle, the transmitter repeats P(DQ) in the second retransmission cycle. If DQ is re-covered by P(DQ) but Dj has yet been recovered, the time slot corresponding to DQ during the second re-transmission cycle will be used to send P(Dj) as illustrated in FIGS. 3 and 4. Since Dj may be re-covered by a P(Dj) in the middle of the first retrans-mission eycle, the transmitter may stop retransmission in the middle of the second retransmission cyele.
Henee, as shown, the second retransmission cycle may eonsist of less than N parity blocks, i.e., it may not be a eomplete cycle. If the receiver fails to recover Dj from the parity blocks P(Dj) transmitted in the first retransmission cycle, the transmitter will com-plete the second retransmission cycle.

If the receiver fails to recover Dj from the first parity block P(Dj) in the second retransmission cycle, it will instruet the transmitter to start the third retransmission eyele. The parity blocks are seleeted in the same manner as that in the second retransmission cycle. The third retransmission cycle will be ter-minated as soon as the transmitter is notified of the recovery of Dj. Retransmission cycles continue until the first erroneously reeeived data bloek Dj has been recovered.

The data recovery procedure can be explained by letting L be the number of consecutive data blocks which are successfully received between the first erroneously received data block Dj and the next S erroneously received data block Dj+L+l with 0<L<N-l.
If l+N-l, then all the N-l data blocks following Dj are successfully received. Therefore, the first L+l consecutive parity blocks are P(Dj ) during the first retransmission cycle.

When the first parity block P(Dj) is received, its validity is checked. If P(Dj ) is successfully re-ceived, the data block Dj is recovered from P(Dj) by an inversion process. If P(Dj) is detected in error, then P(Dj) and the erroneous data block Dj (stored in the buffer) together are used for correcting errors in Dj based on the decoding rule of Cl. Once Dj is recovered, the receiver outputs the L+l good data blocks, Dj ,Dj+l, . . . ,Dj+L. At the same time, the re-ceiver sends L+l consecutive ACK's to the transmitter and ignores the next L incoming parity blocks which are P(Dj ) . Upon receiving these ACK's, the trans-mitter starts to send new data blocks Dj+N,Dj+N+L,....
as shown in FIG. 2.

If the receiver fails to recover the data block Dj fxom the first received parity block P.(Dj ), a NAK ( j ) is again sent to the transmitter. If L>1, the next xeceived parity block is also P(Dj ) . Upon receiving this parity block, the receiver repeats the same pro-cess to recoverDi. If Dj is successfully recovered from the second P(Dj), the receiver outputs data ~locks Dj ,Dj*l, . . . ,Dj+L. This time, the receiver issues L consecutive ACK' s to the transmitter and ignores the next L-l incoming parity blocks which are P (Dj ) . Upon receiving these L-ACK's, the transmitter starts to sends new data blocks, Dj+N,Dj+N+l,....

YO979-0~6 ~ ~ ~r~

Dj+N+L 1' as shown in FIG. 3. Note that, since the receiver fails to recover Dj from the first received parity block P(Dj), the transmitter has repeated the parity block P~Dj) before sending new data blocks Dj+N,Dj+N+l,.... When this parity is received, it will be ignored and an ACK will be sent to the trans-mitter. If the half-rate code ~1 is powerful enough, Dj will be recovered by the first received parity block P(Dj) with high probability. If the first received parity block P(Dj) fails to recover Dj, the second received parity block P(Dj) should be able to recover Dj. The probability that both the first and second P(Dj) fail to recover Dj is very small. The above process continues until Dj is recovered by a received parity block P(Dj).

If the receiver fails to recover Dj when the parity block P~Dj+L+l) for the next erroneous data block Dj+L+l arrives, the receiver swithces to recover Dj~L+l from P(Dj+L+l). No matter whether Dj+L+1 is recovered or not, if the next received parity block is P(Dj), the receiver again attempts to recover Dj, as shown in FIG. 4. As soon as Dj has been successfully recovered, the receiver starts to output good data blocks Dj,Dj+l,... in consecutive order and the trans-mitter is instructed to send new data blocks untilthe next NAK is received.
.
Using SPREC technigue of the present invention, the receiver will not output any data block and the trans-mitter will not be instructed to send new data blocks until the earliest erroneous data block D; has been recovered. Unless the channel error rate is very high, the erroneous data blocks, especially the earliest one, will be recovered in the first retrans-mission cycle with high probability. If a NAK(Q) for the data block DQ is received while the transmitter ~0979-066 is sending new data blocks, the transmitter will initiate the first transmission cycle for DQ no matter whether .he parity block P(D~) has been transmitted before or not.

Referring to FIG. 5, there is shown a system block dia~ram of the SPREC transmitter. As described above, a data block consists of information digits and parity digits based on the information block D and the (n,k) code C0. An information source 10 provides the in-formation digits via a gate 12 as information digitsDi to both a C0 encoder 14 and a Cl parity encoder 16.
As mentioned above, the code C0 is a high-rate (n,k) cyclic or shortened cyclic code which is used for error detection only. The second code Cl is an in-vertible half-rate (2k,k) code which is capable of correcting t or fewer transmission errors and simul-taneously detecting d (d>t) or fewer errors. The encoder 14 encodes the data block Di by C0 and then transmits the encoded data on channel 18.

The parity encoder 16 encodes the data by the code C
to form its parity block P(Di) and stores it in a buffer 20 where its address is determined by a sequenced number.

A control circuit 22 provides data transmission con-25 trol for gating information through the gate 12 by ~
means of control signals on line 24, and also pro-vides parity retransmission control on line 26 to a buffer 20 a parity block selection circuit 28 and a gate 30. Acknowledgements are received on feedback line 32 from the channel 18 and stored in a feedback buffer 34. If the feedback acknowledgements are positive (ACK) or negative (NAK), the control circuit will operate accordingly. More specifically, during Yos 7 ~ - ~ 6 6 .

normal transmission, if a positive acknowledgement is received, a new data block is formed from the infor-mation source 10, a parity code Cl is stored in the buffer 20 and the data block is transmitted via the C0 encoder 14. During the recovery phase if a posi-tive acknowledgement (ACK) has been received on line 32, the control circuit 22 causes the parity block selection circuit 28 to transmit from the buffer 20 the parity block of the earliest data block which has not been positively acknowledged. This parity block P(Dj) is outputted on line 36 through the gate 30 to the encoder 14 which sends the parity block denoted by P(D),Q.

On the other hand, if a negative acknowledgement (NAK) of a not yet positively acknowledged block is received, the control circuit 22 and the parity block selection circuit 28 employ a sequence number or address to obtain from the buffer 20 the parity block correspond-ing to the negatively acknowledged data.

In this system, the receiver will not output any data block and the transmitter will not be instructed to send new data blocks until the earliest erroneous data block Dj has been recovered. Thus, the positive acknowledgement will cause transmission of the parity block of the earliest data block which has not been positively acknowledged.

Referring to FIG. 6, there is shown a system flow diagram for the transmitter shown in FIG. 5. Here, the operation of the transmitter shown and described 3~ with reference to FIG. 5 is disclosed in a flow dia-gram form. ~e~erring to FIG. 7, there is shown a system block diagram of the SPREC receiver. FIG. ~
is a flow diagram corresponding to the operation of the receiver shown in FIG. 7. The data or parity YO979-066 ~ 7 ~

~lock is received on the transmission channel 18 in an error detection circuit C0 indicated by ~umeral 38 which checks the validity of the data block D, Q by using the parity digits which, as mentioned above, are formed based on the information block D and the code C0. When a received data block is detected in error by detector 38, the receiver requests for a retrans-mission by means of the control circuit 40. A detector 42 determines if the received data is a data block or a parity block and, the erroneous information block D is passed through a multiplexor 44 and stored in a buffer 46. As described above, the retransmission is not the original data block (D,Q), but a parity block of n digits, denoted by P(D),Q(l), which consists of the parts P(D), and Q(l).

On the other hand, when a parity block is received, the detector 38 uses the part Q(l) to check the valid-ity P(D). If no errors are detected in detector 38, the parity digits P(D) is used to recover the original data block D by means of an inversion circuit 50 once the detector 42 indicates a parity block. If the data Dl has been recovered, the parity P(D) is dis-carded when so indicated by a detector 48 and the control circuit 40 sends an acknowledge (ACK) to the transmitter.

If the presence of errors is detected by detector 38, then the parity digits P(D) and the previously re-ceived data block D are together used for error correction, as shown by lines 54 and 56 leading into a decoder 58 for Cl. This error correction is based on the decoding rule of code Cl and is conventional decoding as indicated above. Again, where the errors are detected in the parity block, the corresponding data block is retrieved from the buffer 46 by means of the control circuit 40 and a data block selection yo~,9-066 circuit 60 and the Cl decoder 58 attempts to correct the errors in the data and parity blocks. Retrans-missions continue until the original data is re-covered either by inversion or by decoding where it is placed via line 62 through a multiple~or 64 for subsequent use by a user 66 or for storage at 68 at a proper location in the buffer 46.

When a detector 70 indicates that the data Dj has been recovered, a signal is-provided on line 72 to the control circuit 40 to provide a positive acknow-ledgement (ACK).

If no errors were detected in the data block or the parity block, or if error correction has been success-ful, then a form of AND gate 74 causes the data block to be stored via the store 68 into the receiver buffer 46 if there is any earlier data that has not been re-covered. On the other hand, if all earlier data has been recovered, then the data block is passed on to the user 66.

The flow diagram of FIG. 8 shows the operations of the control circuit 40 with the receiver elements shown in FIG. 7.

In normal operation, the transmitter, shown in FIG. 5, sends data blocks via channel line 18 to the receiver continuously. The receiver, shown in FIG. 7, checks the validity of each incoming data block based on error detecting code C0, and sends acknowledgement to the transmitter of each received block. When a received data block is detected in error, a negative acknowledgement (NAK) is sent to the transmitter and the erroneous data block is saved in buffer 46. The receiver proceeds to check the validity of the subsequent N-l received data blocks YO9,9-Q~
3~ ~ r7 ~ ~ ;1~

and stores them in buffer ~6.

Upon receiving the NAK, the transmitter backs up N
blocks to Dj (i.e., Go-Back-N) and starts retrans-missions. The retransmissions are not the N original data blocks, but the parity block P(Dj) and N-l properly selected parity blocks taken by the parity block selectiGn circuit 28. The selection is carried out in the following manner: If the data block Dj+
is successfully received by the receiver, the trans~
mitter does not send P(Dj+l) but uses the time slot to repeat the parity block P(Dj). However, if the data block Dj+l is not successfully received, the transmitter sends the parity block P(Dj+l). This parity block selection process continues for the sub-sequent N-2 retransmissions.

~et L be the number of consecutive data blocks which are successfully received between the first erroneous-ly received data block Dj and the next erroneously received data block Dj+L+l with O<L<N-l (if L=N-l, then all th~ N-l data blocks following Dj are success-fully received). Therefore, the first L+l consecutive parity blocks are P(Dj) during the first retrans~
mission cycle (sending P(D;) and N-l properly selec--ted parity blocks described above is referred to a~
retransmission cycle). If p is the probability that a block will be received successfully, then there are, on average, p~ parity blocks P(Dj) during the first retransmission cycle.

When the first parity block is received, its validity is checked. If successfully receive~, the data block Dj is recovered from the parity block by in-verter 50. If the parity bloc~ is detected in error, then it and the erroneous data block stored ln the buffer 46 together are used for correcting errors in ~ ~'.3 the data block by the Cl decoder 58. Once Dj is re-covered, the receiver outputs the L+l good data blocks Dj, Dj+l,...,Dj+L. At the same time, the receiver sends L+l consecutive ACK's (positive acknowledgements) to the transmitter and ignores the ne~t L incoming parity blocks which are P(Dj).
Upon receiving these ACK's, the transmitter starts to send new data blocks Dj+N, Dj+N+l, Dj+N+L,....

If the receiver fails to recover the data block D.
from the first received parity block P(Dj~, a NAK(j) is again sent to the transmitter. If L>l, the next received parity block is also P(Dj). Upon receiving this parity block, the receiver repeats the same pro-cess to recover Dj. If Dj is successfully recovered from the second P(Dj), the receiver outputs data ~locks Dj, Dj+l, ..., Dj+L. This time, the receiver issues L consecutive ACK's to the transmitter and ignores the next L-l incoming parity blocks which are P(Dj). Upon receiving these L ACK'a, the trans-mitter starts to send new data blocks Dj+N, Dj+N+l, ' j+N+L-l''''' If the receiver fails to reeover Dj when the parity - block P(Dj+L+l) for the next erroneous data block Dj+L+l arrives, the receiver switehes to recover Dj+L+l from P(Dj+L+l), No matter whether Dj+L+l is recovered or not, if the next received parity block is P(Dj), the receiver again attempts to re-eover Dj. As soon as Dj has been suecessfully re-eovered, the receiver starts to output good data blocks Dj, Dj+l, ... in eonsecutive order and the transmitter is instructed to send new data bloc~s until the next ~AK is received.

Referring to FIG. 9, the contrcl program of the transmitter examines the acknowledgement received, if any, as shown in flow block 80. If a negative acknowledgement of a data block not yet positively acknowledged is received at 82, the control program in control circuit 22 shown in FIG. 5 and the parity block selection circuit 28 cause selection at 84 of the corresponding parity block to be next transmitted.
A header is formed and the parity block is encoded by encoder C0 shown in FIG. 5 and then transmitted. Where there is no negative acknowledgement of a data block not yet positively acknowledged as indicated on line 86 in FIG. 9, the positive acknowledgement, if any, is recorded at 88. If the receiver buffer is full at 90, the control program in control circuit 22 selects and removes from the buffer the parity block of the earliest data block which has not been positively acknowledged. A header is formed and the parity block is encoded by C0 and transmitted as indicated by step 92. On the other hand, if the receiver buffer is not full at 90, then a new data block is formed at 94 from the information source, a header is formed and a sequence number is assigned thereto, and the next sequence number count is updated. The new data block is encoded at 96 by C0, and then transmitted.

~~. I' ~0979-066 ~3 Also, the data block is encoded at 98 by Cl to form its parity bloc~ and the parity block is stored in the buffer. The process is repeated as shown at block 80 in FIG. 9.

Referring to FIGS. 10~ and lOB, the control program of the receiver receives the data block or parity block at 100 and provides the status of each block in the input buffer. A sequence number of each block in the input buffer is provided. If there has been retransmission of a positively acknowledged data block at 102, a positive acknowledgement is constructed and transmitted at 104 on line 106. If there has not been retransmission of a positively acknowledged data block at 102, then an indication of a data block or parity block is provided at 108.

As mentioned above, the receiver checks the validity of each incoming data block based on the error detect-ing code C0 as shown at 110, and when a received data block is detected in error at 112, a negative acknow-ledgement (NAK) is sent to the transmitter at 115.If the data block is correctly received at 114, or if error correction has been successful at 116, 118, and 120, the data block is stored in the receiver buffer at 122 and a positive acknowledgement is constructed and transmitted at 124. If there is any earlier data blocks which have not been recovered at 126, the negative acknowledgement is sent to the transmitter.
If there are no earlier data blocks which ha~e not been recovered, then all correct data blocks are out-putted at 128 up to the next erroneous one, if any.
\

When a first parity block is correctly received at130, the data block is recovered from the parity block ~y inversion at 132, and the data block is stored in the receiver buffer at 122 via line 134.

~'097~-0 ~.~ 3 If the parity block is detected in error at 130, then such erroneous parity block and the erroneous data bloc~ stored in the buffer at 116 are used together for correcting errors in the data block by the Cl decoder. Once the error corrections are successful and the data is recovered, that the receiver stores the good data blocks at 122.

If the receiver fails to recover the data block from the first received parity block at 120, a negative acknowledgement is constructed at 115 and sent to the transmitter as shown by line 136.

The present invention results in a high system throughput over a wide range of channel error rates.
FIGS. llA and llB are data and parity block arrange-ments at a transmitter using the selective parityretransmission scheme (SPR~C) for the purpose of illustrating the recovery period of the transmitter.
Here, it is assumed that the transmitter has infinite number of data blocks to be transmitted. The trans-mitter is said to be in the normal state if the nextN blocks to be transmitted are new data bloc~s 140.
Let TA be an instant at which the transmitter is in the normal state and begins to transmit a data block, say (Dj,Qj). We regard this time instant TA as a regeneration point if the transmission of (Dj,Qj) becomes unsuccessful. That is, if errors are detected in (Dj,Qj~ at the receiver. From one regeneration point to the next regeneration point, the transmitter goes through a sequence of events. The stochastic description of the events ~etween a given pair of adjacent regeneration points is identical and in-dependent of the description of the events between every other pair of adjacent xegeneration points.

YOs79-066 Let X be the total number of blocks, data or parity, transmitted during the interval between two adjacent regeneration points. Let Y be the number of data blocks that are accepted by the receiver and delivered to the user during the same interval. Clearly, X and Y are two random variables. Let E[X] and E[Y] be the expectations of X and Y respectively. Then, the throughput of a communication system using the SPREC
scheme of the present invention is defined by nSPREC Q ~~~r. (4) Let Z be the total number of parity blocks trans-mitted during the interval between the instant indi-cated at 142 at which the first NAK(j) for the erroneously received data block Dj is received and the instant indicated at 144 at which the first parity block PtDj) that will recover Dj at the re-ceiver is to be transmitted as illustrated in FIGS.
llA and 11B. Z is a random variable and is referred to as the recovery period for the erroneously re-ceived data block. The throughput ~SPREC depends onE[Z]. FIG. llA shows the recovery period Z where O<~<N. Fig. 11B shows the recovery period Z where N<Z.

Let p be the probability that a block, data or parity, will be successfully received. Let ql be the conditional probability that a data block DQ is successfully recovered from the first received parity block P(DQ) either by inversion or by decoding based on the code Cl, given that errors are detected in the received data block DQ. Let g2 be the condi-tional probability that a data block DQ is success-fully recovered from the second received parity block P(DQ), given that errors are detected in DQ ~nd that ~0979-0~6 r7 5L ~ i the first received parity block P(DQ) fails to recover DQ but detects the presence of errors in DQ and the first P(DQ). Tt is clear that ql~q2~P' Now, we define the following parameters:

~ = p + (L -- p)ql, p) (1 - Pql) (1 - ql) (1 q2) 1 -- p _ ( ql) (q2 - P) 1 -- p Then the average recovery period E~Z] for an erroneous-ly received data block can be upper bounded in Equation 5 as follows:

E[Z] < Zl + ~ (N ~) (5) where

2{l-[l+(N-l)p2] (1_p2~N-l}

+ p~ +(N-l)p](l-p)N 1}, Rl ql + ~ [1- (l-p2)N 1] + ~ [1- (1 p~N-l]

R2 = 1 - ~, qp{l-[l+(N-l)pql~ (l-pql)N 1}, Let A denote the upper bound on E~Z] given by Equation (5). Then the throughput of SPREC can be lower 'i~9/3-0~.6 bounded as below:

q 1 ~SPREC - l _ q ~N 1+(1_~)(N+~) (6) It is noted that a more detailed derivation of the through~ut equations and their analysis is described in I~* ~esearch Report RC 7591 by the co-inventors S. Lin and P. S. Yu in their report entitled, "SPREC -An Effective Hybrid-ARQ Scheme" published in December, 1980.

~rom the above, it can be seen that, in order to main-tain high throughput, must be kept large and A must be kept small. The parameter ~ depends on probabili-ties p and ql. When the channel error rate is very low, p is close to one and ~ ' p. However, p decreases rapidly as the channel error rate increases. The probability ql depends on the error-correcting capability of the code Cl. When p becomes small, if ql maintains large, the parameter ~ would maintain large. The upper bound ~ on the average recovery period depends on p, ql, and q2. If p is large, A
would be small. If p becomes small but ql and q2 maintain large, ~ will be maintained small.

The lower bound on the throughput nspREC given by Equation (6) applies to channels with either random or/an burst errors. It can be evaluated if p, ql, and q2 are known. In order to see the significant gain in the throughput of the SPREC over the Go-Back-N
AR~, we assume that the channel is a binary symmetric channel (BSC) with bit error rate ~ (or the transition probability). Since the probability of an undetected error of C0 is very small for sufficient large n-k [see Equation (3)], we have * REGISTERED TRADE MARK

p _ p = (l-)n (7) which is simply the probability that a block is re-ceived correctly. The probability ql is given below:
1 ~ t (2k) i 2k i 2 -2(1-~) [(1-) +~ (1-) ~}. (8) To derive an exact expression for the conditional probability q2 is tedious, however, a lower bound on q2 can be obtained easily. This lower bound is given below:
t-l q2 - P + (l-p)~~l-g2~ i~l fiSt-i(l fo St-i) (9) where f = ( ) fi(l_ i Sj = .~ fi -1=l Using the probabilities p, q1, and ~2 given above, the lower bound on the throughput nspREc given by Equation (6) can be evaluated. Using a lower bound on q2 will increase the parameter ~ and thus will decrease the lower bound on nSPREC' For ~arious parameters, n, k, t, and N, the throughput nspREc of SPREC and the throughput nGBN of the conventional Go-Back-N ARQ are plotted in FIGS. 12 and 13 where simulation results are also included. For all the cases, the number of parity digits of C0 is chosen ~097~-06G

to be 24 as shown below.

FIG. 12 shows a lower bound throughput of the SPR~C
using a (525,500) cyclic code as CO and a tlO00,500) block code as Cl with N=128 and t=O, 10 and 20 the symbols ~, O and X are simulation results. The throughputs of the conventional Go-Back-N ARQ and Selective-Repeat ARQ are also included for comparison.

FIG. 13 shows a lower bound throughput of the SPREC
using a (2024,2000) code as CO and a (4000,2000) code as Cl with N=128 and t=O and 40 where the symbols ~ and X are simulation results. The throughputs of the conventional Go-Back-N ARQ and Selective-Repeat ARQ are also included for comparison.

Two block codes are used for Cl in the FIGS. 12 and 13, respectively. One is a (1000,500) shortened BCH code Cl which is obtained from the (1023,523) BCH code by deleting 23 information digits. This is described by W. W. Peterson and E. J. Weldon, Jr. in their textbook entitled, "Error-Correcting Codes", 2nd Edition, published by MIT Press, 1972. The shortened BCH code is also described by S. Lin in his textbook - entitled, "An Introduction to Error-Correcting Codes", published by Prentice-Hill, 1970. The shortened BCH
code is invertible and its minimum distance, dmin, is at least 111. This code is used to correct t for fewer errors and simultaneously to detect dld>t) or fewer errors where t+d<lll. If the code is used for correcting t=10 or fewer errors, it is capable of detecting 100 or fewer errors over a block of 1000 digits. The other bloc~ code used for Cl is a (4000,2000) code (FIG. 13) which is obtained by interleaving the (lQ23,523) BCH code with degree 4 and then deleting g2 information digits. This code is also invertible and has minimum distance at least YO9/9-d66 ~ ,7i~L '1 111. Therefore, in plotting the lower bound on throughput of the SPREC, we use the following two pairs of C0 and Cl:

FIG. 12 FIG. 13 .

k=500 k=2000 Co Cl Co Cl (n,k) (2k,k) - (n,k) (2k,k) (524,500) (1000,500) (2024,2000) (4000,2000) where C0 is a shortened cyclic code with 24 parity digits.

From the throughput plots shown in FIGS. 12 and 13, it can be seen that the SPREC provides significant gain in throughput over the conventional Go-Back-N ARQ
for various n, k, t, where N=128. The throughput of the Go-Back-N ARQ falls rapidly to zero as the bit error rate increases. However, the throughput of SPREC maintains at least 50~ as long as the bit error rate ~ < 2k. Even for t=0, the SPREC provides significant improvement in throughput over the Go-Back-N ARQ .

While the invention has been particularly shown and described with reference to preferred embodiments thereof, it will be ùnderstood by those skilled in the art that the foregoing and other changes in form and details may be made therein without departing from the spirit and scope of the invention.

Claims (18)

The embodiments of the invention in which an exclusive property or privilege is claimed are defined as follows:

1. A method for controlling transmission errors in digital data communication systems, comprising:

sending, at a transmitter, data blocks with parity digits for error detection, detecting, at a receiver, the presence of errors in received data blocks, sending a negative acknowledgement to said transmitter, and storing the erroneous data block, and retransmitting properly selected parity blocks, and not the erroneously received data block, to said receiver for correcting errors in said erroneously received data block.

2. Method as recited in Claim 1, wherein said parity blocks are formed based on a half-rate invertible code in which only the parity digits of a code word are used to determine, by an inversion pro-cess, the corresponding information digits.

3. Method as recited in Claim 1, wherein said transmitter, upon receiving a negative acknow-ledgement, retransmits a parity block of n digits consisting of a first part of k parity digits based on both the original k information digits and a half-rate invertible code C1, and a second part of n-k parity digits based on both the k information digits and a (n,k) high rate cyclic code C0.

4. Method as recited in Claim 1, wherein said re-ceiver, upon detecting an erroneously received data block, stores said data block which is subsequently used with its corresponding parity block to recover the correct data block.

5. Method as recited in Claim 1, wherein said data blocks are checked by the receiver using a high rate cyclic code C0 for error detection.

6. Method as recited in Claim 1, wherein said parity blocks are based on an invertible half-rate (2k,k) code C1.

7. Method as recited in Claim 1, wherein each of said parity blocks consists of a first part P(D) comprising k parity digits which are formed based on the original k information digits D
and a half-rate (2k,k) invertible code C1, and a second part Q(1) comprising (n-k) parity digits checking on P(D).

8. Method as recited in Claim 7, wherein said re-ceiver uses said second part Q(1) of a received parity block to check the validity of its first part P(D).

9. Method as recited in Claim 8, wherein said re-ceiver employs said first part P(D) of a re-ceived parity block to recover the original data block.

10. Method as recited in Claim 1, wherein said transmitter, upon receiving a negative acknow-ledgement, stops transmitting new data blocks and starts retransmitting properly selected parity blocks until said transmitter is noti-fied by said receiver that the information digits of the erroneously received data block has been recovered.

11. Method as recited in Claim 10, wherein said transmitter is instructed to send new data blocks after the earliest received erroneous data block is corrected and recovered at the receiver.

12. A method for controlling transmission errors in digital data communication systems, comprising:

sending, from a transmitter, data blocks which are acknowledged at a receiver, when the presence of errors in one of said data blocks is detected, storing said erroneously received data block at said receiver, retransmitting from said transmitter a parity block having error correcting capability for the data block stored in said buffer, whereby the correct information digits in said data block can be recovered from the parity digits in said parity block through an inver-sion operation.

13. A system for controlling transmission errors in a digital data communication system, com-prising:

a transmitter for sending data blocks consist-ing of information digits and parity digits for error detection, first encoder means for en-coding said data blocks, second encoder means for encoding the data blocks to form associated parity blocks, means for storing said parity blocks, a parity block selection circuit for obtaining from said parity block storing means that parity block corresponding to 2 negatively acknowledged data block, and control gating means responsive to a negative acknowledgement of a data block from a receiver for retrans-mitting a selected parity block corresponding to said negatively acknowledged data block, and a receiver including an error detection means for checking the validity of said received data blocks, control means responsive to said error detection means for sending a positive or nega-tive acknowledgement to said transmitter, a data block buffer for storing data blocks which are detected as erroneous, decoding means for recovering the correct data block from the parity digits in said parity block selected for retransmission by said transmitter, said decoding means including both means for select-ing from said data block buffer the data block to be corrected, and a decoder having inversion means for recovering the original data block using said parity digits and said selected data block.

14. System as recited in Claim 13, wherein said first encoder means at said transmitter employs a high-rate cyclic code which is used for error detection for encoding said data blocks.

15. System as recited in Claim 13, wherein said second encoder means at said transmitter employs an invertible half-rate code to encode said data blocks to form associated parity blocks.

16. System as recited in Claim 13, further com-prising, at the receiver, means for detecting errors in said parity blocks received from said transmitter.

17. System as recited in Claim 13, wherein said de-coding means in said receiver employs an in-vertible half-rate code.

18. System as recited in Claim 13, wherein said error detection means, in said receiver for checking the validity of received data blocks, employs a high rate cyclic code which is used for error detection.