found to offer higher operating frequencies up to 3.2 GHz at a suitable supply voltage of 2.0V consuming 24mW. A supply voltage of only 1.2 V is sufficient to operate at 2.0 GHz dissi- pating only 4.6mW with a 1 p F capacitive load. The process and design technology demonstrated here will enable a signifi- cant reduction of power dissipation, cost, and development time for a high-speed system, such as a multigigabit per second communication system.
Acknowledgment: The authors would like to thank Drs. N. Ieda, T. Sakai, and K. Imai for their continuous encour- agement and support. They also wish to thank Dr. Y. Yama- mot0 and K. Kanisawa for their helpful comments.
Y. KADO 2nd August 1990
Y. OKAZAKI M. SUZUKI T. KOBAYASHI N T T LSI Laboratories
3-1 Morinosata, Wakamiya, Atsugi-shi Kanagawa Pref., 243-01, Japan References
MIYAKE, M., KOBAYASHI, T., and OKAZAKI, Y . : 'Subquarter- micrometer gate-length p-channel and n-channel MOSFET's with extremely shallow source-drain junctions', IEEE Trans., 1989, ED-36, pp. 392-398
OKAZAKI, Y., KOBAYASHI, T., MIYAKE, M., MATSUDA, T., SAKUMA, K., KAWAI, Y., TAKAHASHI, M., and KANISAWA, K . : 'High-performance subquarter-micrometer gate CMOS technology', IEEE Electron. Device Lett., 1990, EDL-11, pp. 134-136
J., and MICHEJDA, J. A.: 'Multigigahertz CMOS dual-modulus pre-
scalar IC', IEEE J . Solid State Circuits, 1988, SC-23, pp. 1189- 1194
M., IMAI, K., and KADO, Y.: 'New well structure for deep submicron CMOS-BiCMOS using thin epitaxy over buried layer and trench isolation'. Symp. VLSI Tech. Digest, 1990, pp. 83-84
KAKUMU, M., KINUGAWA, M., and "IMOTO, K.: 'Choice of power- supply voltage for half-micrometer and lower submicrometre CMOS devices', IEEE Trans., 1990, ED-37, pp. 1334-1342 CONG, H-I., ANDREWS, J. M., BOULIN, D . M., FANG, S. C., HILJENIUS, S.
OKAZAKI, Y., KOBAYASHI, T., KONAKA, S., MORIMOTO, T., TAKAHASHI,
SOME CONTINUOUS MULTIDESTINATION
ARQ SCHEMES FOR HIGH ERROR RATE
CONDITIONS
Indexing terms: Telecommunications, Data transmission, Errors
A class of relatively simple continuous ARQ schemes with repeated transmissions suitable for multidestination commu- nications under high error rate conditions is studied. In real
applications, one can optimise the throughput performance by choosing the best scheme along those investigated with respect to the round-trip delay, the number of receivers and the block error probability. Results show that the optimal scheme may offer a far better performance than Morris's scheme for multidestination environments.
Introduction: The increasing applications of point-to-multi- point communication over a broadcast link such as tele- conferencing and file distribution has prompted new efforts in the research of multidestination automatic repeat request (MARQ) technique^.'-^ The three most popular ARQ schemes in use for point-to-point communications, namely the stop-and-wait (SW), the go-back-N (GBN) and the selective- repeat (SR) schemes, have been recently evaluated for multi- destination environments. The multidestination SR (MSR) ARQ scheme clearly offers the best throughput perf~rmance.~ The complexity of buffering for resequencing makes it imprac- tical for real applications. The multidestination SW (MSW)
and the multidestination GBN (MGBN) schemes are ineffi- cient for high error rate conditions, especially when the round-trip delay and the number of receivers are large.
Three variant MARQ schemes for high error rate channels were evaluated in Reference 7. The three schemes can be con- sidered as extended versions of Moeneclaey and Bruneel's; Sastry's' and Morris's6 schemes for multidestination environ- ments. It was found that Morris' scheme has the best through- put performance among the three. Moeneclaey and Bruneel's scheme is the simplest to implement and offers almost the same throughput performance as Morris's scheme does under very high error rate conditions. In the letter we extend the results to a class of continuous MARQ schemes.
Investigated A R Q schemes: The operation of the investigated continuous ARQ schemes can be described as follows. A chunk of m ( m 2 1) or fewer copies of each data block are transmitted contiguously to the receivers. At each receiver, an error detection procedure is performed on each received copy. A positive acknowledgment (ACK) or a negative acknowl- edgment (NAK) is sent back to the transmitter according to whether the copy is received successfully or erroneously. The data block is considered to be successfully delivered as long as each receiver receives successfully at least one of the transmit- ted copies. If at least one receiver receives all the m copies with error, then, just as in the go-back-N ARQ scheme, the trans- mitter goes back to that data block. The same data block is retransmitted continuously until at least an ACK from each of the receivers which requested the transmitter to retransmit is received.
Consider the transmission of a particular data block. Assume that the round-trip delay between the transmitter and any receiver is the same, i.e., N blocks. As a result, if m I N, then all the m copies have to be transmitted before any response of the data block arrives at the transmitter. If m > N, then the transmitter may receive an ACK for the data block from each receiver before all the m copies are transmitted. When this occurs, the transmitter will start transmitting the next data block rather than continuing transmitting the rest of the copies. Therefore, the phrase 'or fewer' was used in the description of the operation of the investigated ARQ schemes.
Throughput performance: The system in our study consists of one transmitter and K receivers. For simplicity, transmission errors between copies of data blocks occur independently at each receiver and the feedback channel is assumed to be error- free. Let P, denote the probability that a copy of a data block
is received successfully by one receiver.
Consider a particular value of m. Let q(m) denote the throughput efficiency. Clearly, if U m ) represents the average number of transmissions required to successfully deliver a data block, then we have q(m) = l/L(m). It is therefore SUB-
cient to compute the value of L(m). Three cases are considered
separately.
Case 1. m = CO: For this case, a data block is transmitted
continuously until the transmitter learns that each of the receivers receives at least one copy successfully. This scheme was actually evaluated in Reference 7. The result is
Urn)
= S(P,, K )+
N - 1 (1)where
represents the average number of copies transmitted until all K receivers receive at least one copy successfully.
Case 2. I I m 5 N : For this case, all the m copies are trans- mitted before the transmitter receives any response. Therefore, U m ) can be computed recursively by
Case 3. N < rn
<
CO: When rn > N, the transmitter mayreceive an ACK from each receiver before all the rn copies are transmitted. Let Q , ( K ) denote the probability that all the K receivers successfully receive the data block in j copies
Q,M
= C1 - (1 -PsY1“
(4)Let RAK) denote the probability that the K receivers receive the data block successfully exactly at the jth copy
Then L(rn) can be evaluated by
m - N rn
+
“il
(:)cl - (1-
PS)”lii = O
x [(l
-
P,)m]K-i[rn+
2(N - 1)+
S(P,, K - i)] (6) Numerical results and discussions: Fig. 1 shows the through- put efficiencies of the MSR, MGBN, Morris’s scheme, and the optimal scheme among the investigated ones against P ,x ;o 8 - U -c 5 0 6 - 3 a -f
..
30 4 - 2 .5
0 0 2 O L 0 6 0 8 1 0 psFig. 1 Throughput efficiency against P, N = 5 ; K = 5
for N = 5 and K = 5. The throughput efficiencies of the MSR, MGBN and Morris’s schemes are equal to l/S(P,, K), l/[l - N
+
NS(P,, K ) ] and l/[(N-
1x1-
P:)+
S(P,, K)], respectively. One can see that the optimal scheme provides a significant improvement in throughput efficiency over Morris’s scheme for P , E C0.4, 0.91. The percentage of improvement is about 31% at P , = 0.85.Figs. 2 and 3 shows similar results for N = 10, K = 5 and N = 5, K = 20, respectively. Comparing Figs. 1, 2 and 3. one can see that the improvement is more significant when N and/or K are larger. In reality, the improvement at P , = 0.85 is about 101% and 14% for N = 10, K = 5 and N = 5,
K = 20, respectively.
0 0 2 0 4 0 6 0 8 1 0
ps 1811121
Fig. 2 Throughput eflciency against P ,
N = 1 0 ; K = 5
There are several interesting topics in this area that can be studied further. For example, it is worth evaluating the per- formance of various ARQ schemes for multidestination
0 0 2 0 4 0 6 0 8 1 0
ps
1811131
Fig. 3 Throughput efficiency against P, N = 5; K = 20
environments where round-trip delays between the transmitter and different receivers are not the same.
T-H. LEE 18th July 1990
Department of Communication Engineering National Chiao Tung University Hsinchu, Taiwan 30050 Republic of China References
1 GOPAL, I. s., and JAFFE, I. M.: ‘Point-to-multipoint communication over broadcast links’, IEEE Trans., 1984, COM-32, pp. 1034- 1044
2 MASE, K., TAKENNAKA, T., YAMAMOTO, H., and SHINOHARA, M.: ‘Go- back-N ARQ schemes for point-to-multipoint satellite communi- cations’, ibid., 1983, COM-31, pp. 583-590
3 SABNAM, K., and SCHWARTZ, M.: ‘Multidestination protocols for satellite broadcast channels’, ibid., 1985, COM-33, pp. 232-239 4 MOENECLAEY, M., and BRUNEEL, H . : ‘Efficient ARQ scheme for high
error rate channels’, Electron. Lett., 1984,20, pp. 986-987 5 SASTRY, A. R. K.: ‘Improving automatic repeat-request (ARQ) per-
formance on satellite channels under high error rate conditions’, IEEE Trans., 1975, COM-23, pp. 436-439
6 MORRIS, I. M.: ‘On another go-back-N ARQ technique for high error rate conditions’, ibid., 1978, COM-26, pp. 187-189
7 CHO, Y . I., and UN, c . K . : ‘Continuous multidestination ARQ schemes for high error-rate channels’, Electron. Lett., 1988, 24, pp. 694-695
TWO-BAND IIR QUADRATURE M I R R O R FILTER DESIGN
Indexing term: Filters
Two types of two-band IIR quadrature mirror filter structure are proposed. The aliasing distortion and amplitude distor- tion can be exactly cancelled. The Remez exchange algorithm is used iteratively to optimise the filter response, which results in an equal-ripple design.
Introduction: Quadrature mirror filters (QMF) have great applications in sub-band coding systems.’.’ In general, a sub- band coding system suffers from three kinds of distortions: aliasing distortion, amplitude distortion and phase distortion. If a sub-band coding system is free from these three distor- tions, it is called a ‘perfect reconstruction’ system. Theoreti- cally, only FIR filter banks can achieve perfect reconstr~ction.~ Ordinary FIR filters have large transition band. If both a sharp transition band and reasonable stop- band attenuation are required, one must use very high order